Tunisian enterprise AI startup InstaDeep raises $100M from Alpha, BioNTech, Google

A recent survey carried out by CNBC reported that 81% of executives worldwide say AI will play a prominent and critical role in how their businesses operate this year.

Companies are phasing from the first generation of AI, which deals with pattern, text and image recognition, to decision-making AI, which helps them make timely decisions in complex spaces.

InstaDeep, a Tunis and London-based enterprise AI startup that creates decision-making systems for solving real-world problems, has raised $100 million in Series B financing led by Alpha Intelligence Capital and CDIB.

BioNTech (the company behind Pfizer’s COVID-19 vaccine), Chimera Abu Dhabi, Deutsche Bahn’s DB Digital Ventures, Google, G42 and Synergie participated in the round.

InstaDeep was founded by Karim Beguir and Zohra Slim in 2014. The Tunisian startup, headquartered in London with offices in Paris, Tunis, Lagos, Dubai and Cape Town, uses advanced machine learning techniques to bring AI to applications within an enterprise environment.

Beguir, the chief executive officer, on a call with TechCrunch, said the eight-year-old company’s AI and machine learning solves an array of challenges.

They can range from a large shipping company finding ways to efficiently transport thousands of containers to a railway station, with more than 30,000 kilometres of railway, trying to automate scheduling for 10,000 trains. Other examples are the design of advanced therapeutics with silicon and routing components on a printed circuit board.

These types of problems, though in different verticals, have similarities. InstaDeep uses reinforcement learning, a kind of machine learning that helps design optimization strategies and tackles them simultaneously.

In a statement, the company said it is currently working on a moonshot product to automate railway scheduling with Deutsche Bahn. The rail operator is the largest in Europe. 

Two years ago, InstaDeep formed a multi-year strategic collaboration with BioNTech to launch a joint AI innovation lab. The lab’s mandate was to deploy the latest advances in AI and ML to develop novel immunotherapies.

One of its best efforts came in late November when it created an early warning system (EWS) for detecting high-risk SARS-CoV-2 variants. Per a report by FT, this EWS identified more than 90% of World Health Organization (WHO) designated variants on average two months ahead of time and detected Omicron three days before it was classified as a variant of concern by the WHO.

InstaDeep also collaborates with Google’s AI research divisions to create an early detection system for desert locust outbreaks in Africa; it has worked on AI initiatives and has published joint research with DeepMind and Google Research.

A common theme with these partnerships is that all three organizations are investors in InstaDeep’s new financing round.

“With them being our partners and customers, they’ve been able to see firsthand what InstaDeep platform and the team can achieve,” said Beguir. “So we see it as a significant milestone and also sort of a vote of confidence in our capabilities and products that they are investing having worked very closely with us on difficult problems for years.”

Karim Beguir

Karim Beguir (InstaDeep CEO)

Beguir and Slim bootstrapped InstaDeep from 2014 to 2018, pumping revenue from clients back into the business acquiring new talent and expanding. In 2019, the Tunisian startup raised a $7 million Series A round from pan-African private equity firm AfricInvest and New York-based Endeavor Catalyst to scale its systems.

InstaDeep has established itself as a global company using AI to solve complex problems with significant monetary value. For example, building one kilometre of railway costs hundreds of millions of dollars. So, providing an intelligent system — which is one of InstaDeep’s applications — that can optimize train traffic, and manage constraints better, is highly marketable.

With the new funding, the enterprise AI company plans to accelerate the launch of disruptive AI products across biotech, logistics, transportation and electronics manufacturing. Advancing its computing infrastructure, expanding into the U.S. and hiring more talent is also in its use of funds strategy.  

InstaDeep currently has over 170 employees. More than 130 are in AI research, engineering, ML and DevOps departments, while half of the team is based in its African offices: South Africa, Nigeria and Tunisia.

When InstaDeep launched, Africa wasn’t in the picture detailing AI’s contribution to global economic growth. And while that picture hasn’t changed so much, InstaDeep is one of the few African companies, including South Africa’s Aerobotics and hearX Group, trying to change that status quo and give Africa have a say in shaping the future of AI.

“We’ve managed to build a culture of high standards and prove that the talents in Africa are capable of being competitive, working and collaborating with the very best,” said Beguir. “That’s the story we’ve been able to nurture. And today, we’re proud to have a team which is now over multiple countries in Europe, Middle East and Africa, but has some very passionate African AI researchers, engineers making a tangible contribution.”

Beguir mentioned on the call that at the time InstaDeep started with “two laptops, $2,000 and a lot of enthusiasm,” many investors and onlookers within the African tech and AI space doubted the company’s goal to collaborate with the likes of DeepMind and Google.

But if technology has taught us anything, location doesn’t pose a barrier in getting global customers. And this holds more true for AI and deep tech technology as long as companies have access to knowledge, talent with experience and an open AI community.

Beguir, half Tunisian and half French, grew up in the North African country but studied engineering and mathematics in France and the U.S.

After a classical career background, Beguir said he started InstaDeep to prove that African talent could be competitive, make a difference in deep tech and collaborate and compete with the best in the world.

“It is possible to create a globally competitive company with strong African roots, but also well integrated into the world working on genuine deep-tech innovation, and doing things that haven’t been done before,” the CEO said. “That’s been our story so far, and we can’t wait to take it to the next level with our investors and partners and try to have a positive impact on the ecosystems in which we operate and all the partners with whom we work.”

What Kai-Fu Lee-backed AInnovation tells us about China’s smart manufacturing

The enthusiasm to find paying customers for artificial intelligence continues in China. AInnovation, a Chinese computer vision startup backed by Kai-Fu Lee’s Sinovation Ventures and SoftBank, is trying to automate China’s massive manufacturing industry. Merely four-year-old, the startup has filed to go public in Hong Kong, and its prospectus is offering a rare glimpse into the commercial viability of smart manufacturing, which is a key part of China’s industrial blueprint for the next few years.

Throughout the 2010s, computer vision companies like SenseTime and Megvii struck gold by powering China’s public security infrastructure with facial recognition technology. As competition pushes down prices and pressure from U.S. sanctions over surveillance tech rises, China’s early AI upstarts have sought diversification. SenseTime has gone into education. Megvii, also backed by Sinovation Ventures, added warehousing to its offerings.

AInnovation is among a rank of young players in the AI application arena. Co-founded by CEO Xu Hui, who’s a veteran of IBM, SAP and Microsoft in China, the startup derived half of its revenues from manufacturing customers for the nine months ended September 2021, according to its prospectus. Its computer vision modules and customized services are used in scenarios like transporting molten iron, detecting abnormalities on automotive production lines and spotting defects during semiconductor manufacturing.

A third of the firm’s revenues were derived from financial services, with the rest coming from retail, telecommunications and other industries.

A business like AInnovation can’t just hire a team of PhDs to run machine learning models in the lab. It needs to, literally, get its hands dirty, visit customers’ plants, and learn what type of automation generates the best return for steelmakers and clothing manufacturers. As such, the startup has established two joint ventures respectively with its key partners — major steelmaking group CISDI and state-owned construction giant China Railway No. 4.

Detecting screw defects using AInnovation’s computer vision technology. Photo: AInnovation

AInnovation isn’t generating nearly as much income as its smart city predecesosrs, yet. In 2020, the startup raked in 462 million yuan ($73 million) in revenue; SenseTime pocketed 3.4 billion yuan in that year. But AInnovation is growing rapidly. In the nine months ended September 2021, its revenue reached 553 million yuan, exceeding the sum of 2020.

There are challenges though. For one, the startup relies heavily on a few key accounts. Revenue generated from its five largest customers in 2019 and 2020 accounted for around 26% and 31%, respectively.

China’s early AI contestants gathered around facial recognition for a reason — it’s lucrative because it’s mostly a software business. SenseTime’s profit margin, for example, rose from about 57% in 2018 to over 70% in 2020.

AInnovation was once a software-first business as well. Its gross margin stood at 63% in 2018 but plummeted to 31% in 2019 and further to 29% in 2020, because it pivoted from selling mostly software to integrated solutions involving more hardware components, which generally incur more material costs, its prospectus notes. Profitability decreased also because the company “offered competitive pricing” to expand its customer base. In the AI business, data is the fuel.

Both are still unprofitable businesses. AInnovation posted an adjusted net loss of about 160 million yuan ($25 million) in 2019 and 144 million yuan in 2020. SenseTime, in comparison, recorded an adjusted net loss of 1 billion yuan and 878 million yuan during the same periods.

Every vertical within China’s manufacturing industry is easily a multi-billion market opportunity; the question is whether AInnovation can find its way to sustained growth and a healthy business model.

AInnovation has priced its shares at HK$26.30 ($3.38) apiece, the bottom of its marketed range, Bloomberg reported earlier. At this price, the company would raise about $151 million from its IPO in Hong Kong.

AI2 shows off an open, Q&A-focused rival to GPT3

OpenAI’s impressive AI language model GPT-3 has plenty of things going it, but with 175 billion parameters no one would claim it’s particularly streamlined. The Allen Institute for AI (AI2) has demonstrated a model that performs as well or better than GPT-3 on answering questions, but is a tenth the size.

Macaw, AI2’s model, emerged from research being done at the nonprofit into creating an AI that performs at human levels on standardized tests.

“After we got a very high score they moved on to harder questions,” said AI2 head Oren Etzioni. “There’s this paradox where sometimes the questions that are easiest for people are the hardest for machines — and the biggest gap was in common sense.”

For instance, he said, asking “When did Tom Hanks land on the moon?” GPT-3 says 1995, since that’s when the film Apollo 13 came out. But Macaw correctly notes that, as far as we know, Tom Hanks has in fact never been to the moon.

AI2 has published a collection of examples (“demo” is probably too strong a word) that illustrate the difference, with Macaw often providing the better answer to a variety of tricky and not-so-tricky questions.

It’s humorous to compare the two, as sometimes they both miss or give absurd answers at a pretty high rate. Macaw is good at tracking who did what, but is convinced snakes have legs. GPT-3’s answers are confident (perhaps overly so) and readable, even when they amount to conspiracy theories like that COVID-19 is “a new, experimental, and potentially dangerous drug.”

A couple examples:

  • How would you make a marriage stronger?
    Macaw: Communicate
    GPT-3: I would make sure that I was the one who was always right.
  • If a bird didn’t have wings, how would it be affected?
    Macaw: It would be unable to fly
    GPT-3: It would be a bad bird.
  • You poured yourself a glass of cranberry juice, but then absentmindedly, you poured about a teaspoon of grape juice into it. It looks OK. You try sniffing it, but you have a bad cold, so you can’t smell anything. You are very thirsty. So you drink it. What happens next?
    Macaw: You get sick
    GPT-3: You will probably die
  • How does a snake move around?
    Macaw: By slithering on all fours
    GPT-3: Snakes move by using their muscles to push against the ground. They can move in any direction, but they cannot move backwards.

Clearly there’s still a ways to go in creating a question-answering AI that doesn’t trip up on basic questions that a child could answer. But it’s also worth noting that Macaw achieves a similar level of success with a far, far less data-intensive process. Etzioni was clear that this is not meant to be a GPT-3 replacement in any way, just another step in the research going on worldwide to advance the ball on language generation and understanding.

“GPT-3 is amazing, but it only came out 18 months ago, and access is limited,” he said. The capabilities it demonstrated are remarkable, “But we’re learning you can do more with less. Sometimes you have to build something with 175 billion parameters to say, well, maybe we can do this with 10 billion.”

A good question-answering AI isn’t just good for party tricks, but is central to things like voice-powered search. A local model that can answer simple questions quickly and correctly without consulting outside sources is fundamentally valuable, and it’s unlikely your Amazon Echo is going to run GPT-3 — it would be like buying a semi truck to go to the grocery store with. Large scale models will continue to be useful, but pared-down ones will likely be the ones being deployed.

A part of Macaw not on display, but being actively pursued by the AI2 team, is explaining the answer. Why does Macaw think snakes have legs? If it can’t explain that, it’s hard to figure out where the model went wrong. But Etzioni said that this is an interesting and difficult process on its own.

“The problem with explanations is they can be really misleading,” he said. He cited an example where Netflix “explains” why it recommended a show to a viewer — but it’s not the real explanation, which has to do with complex statistical models. People don’t want to hear what’s relevant to the machine but rather to their own mind.

“Our team is building these bona fide explanations,” said Etzioni, noting they had published some work but that it isn’t ready for public consumption.

However, like most stuff AI2 builds, Macaw is open source. If you’re curious about it, the code is here to play with, so go to town.

Meta leaps into the supercomputer game with its AI Research SuperCluster

There’s a global competition to build the biggest, most powerful computers on the planet, and Meta (AKA Facebook) is about to jump into the melee with the “AI Research SuperCluster,” or RSC. Once fully operational it may well sit in the top ten fastest supercomputers in the world, which it will use for the massive number crunching needed for language and computer vision modeling.

Large AI models, of which OpenAI’s GPT-3 is probably the best known, don’t get put together on laptops and desktops; they’re the final product of weeks and months of sustained calculations by high performance computing systems that dwarf even the most cutting-edge gaming rig. And the faster you can complete the training process for a model, the faster you can test it and produce a new and better one. When training times are measured in months, that really matters.

RSC is up and running and the company’s researchers are already putting it to work… with user-generated data, it must be said, though Meta was careful to say that it is encrypted until training time and the whole facility is isolated from the wider internet.

The team that put RSC together is rightly proud at having pulled this off almost entirely remotely — supercomputers are surprisingly physical constructions, with base considerations like heat, cabling, and interconnect affecting performance and design. Exabytes of storage sound big enough digitally, but they actually need to exist somewhere too, on site and accessible at a microsecond’s notice. (Pure Storage is also proud of the setup they put together for this.)

RSC is currently 760 Nvidia DGX A100 systems with a total 6,080 GPUs, which Meta claims should put it approximately in competition with Perlmutter at Lawrence Berkeley National Lab. That’s the fifth most powerful supercomputer in operation right now, according to longtime ranking site Top 500. (#1 is Fugaku in Japan by a long shot, in case you’re wondering.)

That could change as the company continues building out the system. Ultimately they plan for it to be about three times more powerful, which would in theory put it in the running for third place.

There’s arguably a caveat in there. Systems like second-place Summit at Lawrence Livermore National Lab are employed for research purposes where precision is at a premium. If you’re simulating the molecules in a region the Earth’s atmosphere at unprecedented detail levels, you need to take every calculation out to a whole lot of decimal points. And that means those calculations are more computationally expensive.

Meta explained that AI applications don’t require a similar degree of precision, since the results don’t hinge on that thousandth of a percent — inference operations end up producing things like “90% certainty this is a cat,” and if that number were 89% or 91% wouldn’t make a big difference. The difficulty is more about achieving 90% certainty for a million objects or phrases rather than a hundred.

It’s an oversimplification, but the result is that RSC, running TensorFloat-32 math mode, can get more FLOP/s (floating point operations per second) per core than other, more precision-oriented systems. In this case it’s up to 1,895,000 teraFLOP/s or 1.9 exaFLOP/s, more than 4x Fugaku’s. Does that matter? And if so, to whom? If anyone, it might matter to the Top 500 folks, so I’ve asked if they have any input on it. But it doesn’t change the fact that RSC will be among the fastest computers in the world, perhaps the fastest to be operated by a private company for its own purposes.

Klarity lands $18M to read scores of documents so you don’t have to

Reviewing repetitive documents is, well, repetitive, but Klarity believes people don’t have to do all of that and is building an artificial intelligence tool, targeting finance and accounting departments, that turns documents into structured data.

Document automation is not a new concept. There was an original wave of companies working on partial document automation, which still needs a human review, but Ondrej Antos, Klarity’s co-founder and CEO, explained that the full document automation market is still very nascent.

“Partial document automation companies did not achieve much scale due to the limited value of their product,” he said via email. “Full automation has the ability to replace human review for a vast majority of documents — over 85% in Klarity’s case — and with a higher accuracy. This generates a lot of value, not only for large enterprises, but also mid-market companies that have a few hundred documents every month and the market is therefore much larger.”

Antos founded Klarity in 2017 with Nischal Nadhamuni whom he met at MIT. They bonded over Antos’ experience of having to review large amounts of data when he was a corporate lawyer. Nadhamuni was studying Natural Language Processing and thought it could be applied to  understand documents better than humans. In August 2020, the product was launched.

Klarity replaces humans for tasks that require large-scale document review, including accounting order forms, purchase orders and agreements. Instead of having many accountants reading thousands of almost identical documents every month to find non-standard language, Klarity does that, helping the accountants save time and avoid mistakes.

Klarity

An example of Klarity’s document automation. Image Credits: Klarity

Over the last nine months, the company saw its annual recurring revenue grow nine times and over 24 times year over year, prompting Klarity to raise new capital to invest in sales and marketing to scale and continue investing in R&D. It is also currently working with more than 40 enterprise and mid-market customers, including Coupa, Optimizely and 8×8.

Today, the company announced $18 million in a Series A funding round led by Tola Capital. As part of the investment, Sheila Gulati, founder and managing director of Tola Capital, joins Klarity’s board of directors. To date, Klarity has raised just over $20 million.

New investors also participating in the round are Invus Opportunities and a group of individual investors, including executives from its customers 8×8 and Coupa. Existing investors following on include Elad Gil, Daniel Gross, Nat Friedman and Picus Capital.

The company is focused on hiring sales, marketing and engineering. It has 34 employees, up from 14 a year ago. It is also poised to launch new document review automation use cases for deal desk, renewals and procurement teams in late 2022.

“Today, the vast majority of enterprises don’t even realize there is a technological solution to this omnipresent problem,” Antos said. “We will help to educate the market that there is a technical solution to the age-old problem of document review by accounting teams and to continue building a market-leading product.”

Tekever raises $23M for industrial drone technology optimized for maritime surveillance

Industrial drones — the enterprise complement to the unmanned aerial vehicles that consumers own for leisurely use — are taking off in the market, fueled by a new wave of software and hardware technology that improves their battery life, reach and performance, and a growing number of organizations investing in these services to raise their data operations games. Today, a company focused specifically on developing AI for drones for maritime deployments is announcing a round of funding after seeing strong demand for its devices and services.

Tekever, which builds drones with integrated AI specifically tailored for monitoring and detecting activity on water, has raised €20 million (just under $23 million at today’s rates). Ventura Capital led the round with participation from Iberis Capital and a number of unnamed strategic investors from the maritime industry. It will be using the funding to hire more people, and to continue building out its technology.

Tekever — based, fittingly, in historic maritime superpower Lisbon, Portugal — was founded back in 2001 and has only been offering commercial services since 2018. But it has been profitable for some time now and expects to grow at a CAGR of 60% over the next three years. And indeed, this is the company’s first outside funding, made to double down on expanding its technology and selling to a wider set of organizations as the business opportunity grows.

Tekever’s customers include governments and their agencies, which use the company’s services to monitor waters for illegal activity; and private shipping and other seafaring companies, which use the drones to track weather patterns, traffic on the water, and other physical activity that might impact their businesses.

Tekever was started by a team of intelligence and AI specialists, and co-founder and CEO Ricardo Mendes describes it as a vertically-integrated business, where it designs and makes both its drones and the technology that is loaded onto them to monitor and “read” what is going on in the water below, and even predict what might happen next.

A vertically-integrated drone company is not such a rare thing, but what is a little more unusual is the order in which way Tekever built its stack.

“We started from the opposite direction of every other company working in the drone sector,” Mendes joked. The company first set out to build the technology to read its terrain — in its case, bodies of water — and then built drones fit for the purpose of making its software work. That included specific antennae, sensors and power integrated into the body of the aircraft themselves. (This also makes it essentially, at this point, impossible for the software to work on other aircraft.) The software, meanwhile, is architected to work using a mix of edge AI, satellite communications and cloud computing.

Building your own very specialized drone hardware is hard (and expensive). But that was intentional, it turns out. Tekever sells both components but most commonly operates its own fleet and sells drone-based surveillance services to users, branded Atlas, which Mendes described to me as “intelligence as a service.” He said that approach was specifically taken to make its products as widely accessible as possible, since its drones — with wingspans that start at two meters and can be as wide as eight meters, with flying times as long as 20 hours — are too cost-prohibitive for anyone but the very biggest customers.

“The question we set out to answer was, ‘what do you need to do to make this simple and available all over the world, not just to the richest nations?” he said. “The drones are just one part of the chain.”

As an example of how Tekever is used, both the European Maritime Security Agency (EMSA) and the United Kingdom’s Home Office customers, but so are smaller African republics. They variously use the technology to monitor their waters for vessels involved in piracy, drugs, human trafficking, migrant smuggling, pollution, illegal fishing, or infrastructure security threats.

A recent report in the Guardian laid bare how European government agencies are investing millions of euros on drones and other military technology to expand their surveillance of refugee groups, its clear message being that those investments are not deterring illegal migration and are only encouraging vulnerable people to take even more risky routes. Others in the space like Anduril have reaped huge financial rewards arguably on the back of their own controversy. But Tekever’s CEO and founder believes that his company not only fills a specific technical gap in the market, but that its use ensures more good than it does harm.

“When you are talking about vast regions like the ocean there are a lot of unknowns about what is going on,” he said. Typically, organizations have relied on satellite imagery to get pictures of what is going on in the water, but this is not ideal since most satellite imagery is days old by the time it’s seen by a user. “Fishing, smuggling, trafficking, immigration — these are all areas where real-time intelligence required. It’s not just footage, but the beginning of solving the problem. The objective is that you should be able to act before something bad happens,” and, because Tekever is also using predictive analytics, to preview what will come.

“What we are doing is gathering vast amounts of data to solve problems as they occur,” he said, noting that even having an extra five minutes to respond can make a difference because of how fast conditions can change in the water. For the UK’s Home Office, for example, he noted that one priority has been to identify migrant boats in the English Channel to help escort them to shore, to avoid potentially tragic accidents. “The press focuses on migration issue itself but it’s a huge humanitarian issue,” he said.

Going forward, there is a vast amount — a sea, even — of ways that Tekever might develop its technology. Looking at and making sense of bodies of water requires the crunching of vast amounts of data, Mendes said, but that also gives the company a large number of datasets that could be put to use, too. It has yet to be able to read submarine activity, something that today needs lidar and radar on seacraft to identify; but that is an area it’s starting to develop. Another is to identify and classify oil spills, he said.

Right now, Tekever’s focus remains on what Mendes described to me as “the blue economy”, but it is also breaking ground on… ground. Its focus seems to be to continue trying to create new ways of looking at the most complicated terrains. One area he noted it wants to do more in is forest and specifically rainforest. It invested in a Brazilian drone company, Santos Lab, several years ago, giving it a foothold in that part of the world.

“Tekever is a very unconventional UAS company and a market leader with outclass technology, thousands of hours of operational experience, a seasoned leadership team and a phenomenal and profitable business vision in a fast-growing market,” said Mo El Husseiny, managing partner at Ventura Capital, in a statement. “These attributes have made it a flagship investment for Ventura, aligned to our portfolio of disruptors in technology.”

“Tekever is one of the hottest European Deeptech scaleups, and we’re very proud to continue working with the team and helping them disrupt the global market” added Diogo Chalbert Santos, a partner at Iberis Capital. “It’s amazing what Tekever has already achieved as a bootstrapped business and I’d say not even the sky is a limit with this round.” (It seems that Santos can’t resist a pun, an investor after my own heart.)

Fathom raises $4.7M for its AI notetaker

Fathom, a startup that is building an AI notetaker for Zoom, today announced that it has raised a $4.7 million seed round from a range of early-stage investors, including Zoom’s own Apps Fund. Other investors include the likes of early Zoom investors Maven Ventures, Bill Tai and Matt Ocko, early-stage funds Character.vc, Active Capital, Global Founders Capital, Rackhouse.vc and Soma Capital, as well as the CEOs of Reddit, Twitch, Cruise, Mercury, People.ai, Snapdocs and Shogun. That’s a broad coalition of investors, one that’s surely in part driven by the fact that Fathom founder and CEO Richard White also previously built UserVoice.

“It was actually at UserVoice where I got the idea for what’s now turned into Fathom because I was doing a lot of research and customer calls in service of investigating a different product we were building at UserVoice,” White told me. “I think it was kind of January 2020 […] and I did like 300 Zoom calls in that first month. It was kind of crazy. I’m not sure why I signed up to that many research calls. I got really well acquainted to what a terrible experience it is to have to talk to people all day long and type notes at the same time — and then after the call is over, clean up those notes so that they make sense. It’s a very stressful kind of process and you worry you’re going to miss something.”

Image Credits: Fathom

To fix that, White and his team built Fathom to make it easy to not just automatically record and transcribe every Zoom call, but also to quickly generate summaries after the call which can then be emailed to participants or imported into a CRM system, for example. During a call, all you have to do is highlight important parts using the service’s Zoom or desktop app.

While sales teams are some of the obvious users for this kind of service — they tend to have a lot of Zoom calls these days, after all — White argues that there is also a wider market for a service like this and that none of the features are necessarily sales-specific. “Overall, we try to build a platform that is useful for anyone that’s on a decent number of Zoom calls per week,” said White.

Fathom founder and CEO Richard White

Fathom founder and CEO Richard White

To create this product, White took some work that had already been done at UserVoice and then spun that out as a new company — and in the process, the team also made it into Y Combinator. That may seem like a bit of an odd choice for a seasoned entrepreneur, but as White noted, a lot has changed in the last few years and simply having a cohort of like-minded founders that are going through the same processes is a very useful thing to have. “I’ve learned probably as much from the common curriculum at YC as I have from having this peer network of people that are all hustling and doing the same thing,” White noted. He also likened YC to a startup union that can help with fundraising and negotiating with vendors, too.

As of now, Fathom is available for free, with plans to add paid features over time (though the core experience will remain free). As of now, though, support for other platforms like Google Meet or Microsoft Teams is not on the roadmap.

“Zoom has already transformed our lives and the way we do business – now that we live on video, the need and opportunity for Fathom is obvious,” said Jim Scheinman, early Zoom investor and Founding Managing Partner at Maven Ventures. “The Fathom team is incredibly strong and experienced, and they’re solving a key problem that Zoom customers face on a daily basis and will change the way people experience their business calls for the better. Being one of the first deeply integrated Zoom Apps opens the door for massive growth.”

Fathom, of course, isn’t the only company trying to solve this problem. Chorus similarly tries to provide AI-driven solutions for sales calls, for example, while services like Gong focus more on analyzing customer-facing interactions across platforms and Otter mostly focuses on the transcription side of the process.

Caribou grabs $3M to remove the ‘unexpected’ from healthcare cost planning

With almost 4,000 Medicare Advantage plans to choose from, it not only makes choosing one a complex decision, but making a wrong decision could affect the wallet if something unexpected should happen.

Miami-based startup Caribou aims to make that decision easier through its healthcare cost prediction and optimization software packaged into a SaaS business model designed for financial planners so they can advise their clients on the best plan.

The company, led by co-founders Christine Simone and Cory Blumenfeld and founding engineer Giorgio Delgado, gathers data on factors, like utilization, health conditions and medications, and provides financial advisors with a scalable tool to evaluate a client’s healthcare planning needs.

Simone explained that financial advisors don’t often ask clients about their medication costs or health conditions, so some of the pillars the company helps advisors and their clients identify include health plan selection and if you may need long-term care planning — which Simone estimated 70% of people usually do.

Simone and Blumenfeld started Caribou in 2020 after careers in healthcare, where they saw stakeholders not addressing the financial component of care.

“And that burden unfortunately gets placed on the consumer,” Simone told TechCrunch. “Every day we hear about rising healthcare costs and medical bankruptcy, and I’m so excited that we’re empowering consumers to proactively plan for the costs and make smarter decisions using data.”

The company raised $575,000 in early 2021, and today, announced another $3 million in a seed round to bring its total funding to $3.1 million.

The investment was led by Jack and Max Altman, who were joined by Lightspeed Scout Fund, Dash Fund, existing investors Garage Capital and N49P, as well as a group of angel investors, including Plaid CTO Leslie Schrock and Tribe’s Arjun Sethi.

Jack Altman said in a written statement that Caribou is picking up the healthcare planning where employers leave off.

“We’re seeing solutions in the employer space aiming to reduce healthcare spend for employees, companies and payers,” he added. “What happens once people leave those companies or their HR departments no longer have access to that data? Caribou’s positioning through the financial system lens is a great opportunity to reach a different segment of customers and offer them something incredibly valuable.”

Caribou’s software has only been in the market for a few months, but the company is already racking up dozens of customers, including BLB&B Advisors, CapSouth Wealth Management and Jackson Square Capital.

Though it was too early to talk about growth metrics, Simone says growth, especially off of the open enrollment season in the fourth quarter, “was crazy busy in terms of adoption,” and now the company is at a threshold of putting firms on a waitlist. Current customers represent tens of thousands of end clients, and the company is already able to prove that it saved consumers hundreds of thousands of dollars in healthcare costs.

The new funding will be invested in product development to be able to provide more robust financial insights and reach wider-scale distribution. The company also aims to grow its customer success team and double its overall employee headcount by the end of the year.

Caribou’s goal is to focus on the end consumer, so while it is distributing its software through the financial industry, its product roadmap also includes different distribution channels, including technology platforms that consumers could access directly.

“There’s only 39% of Americans that work with a financial advisor, so we also need to be looking at opportunities outside of financial advisors,” Simone said. “We want to be able to distribute our tool like a back-end plug-in embedded into some of these other financial technology platforms to give access to more consumers.”

a16z, Monashees lead new round into inventory discovery startup Inventa

Inventa, a Brazil-based company offering a digital marketplace for small and medium-sized companies to discover and purchase new inventory, raised $20 million in Series A funding.

Andreessen Horowitz and Monashees co-led the round and were joined by Founders Fund, Greenoaks, Greylock, Tiger Global and angel investors Hans Tung and Carlos Gracia from Kavak. Also participating in the round were existing investors Pear VC, NXTP, ONEVC, MAYA Capital and Alter Global.

This fresh infusion of capital comes three months after the company announced $5.5 million in seed funding. And all of that for a company that started in March 2021.

CEO Marcos Salama founded the company with former General Atlantic investor Laura Camargo and former McKinsey data science expert Fernando Carrasco to provide technology, data and credit to Brazilian entrepreneurs.

Salama, who is from Spain, has a background in mechanical engineering and has worked for both McKinsey and Rappi, which is how he made his way to Brazil. While leading Rappi’s groceries business, he worked with retailers and saw how small stores were struggling to access an assortment of goods and credit.

Inventa uses technology to provide an easier purchasing process for small businesses. Inventa’s online platform recommends products based on actual transaction data and provides credit, in 30-, 60- and 90-day increments, to retailers. There is also a supplier side, where they can upload products, manage pricing and see what is selling and what isn’t.

One of the drivers for going after additional funding so soon was that Inventa is growing at over 100% month over month.

“There are 5 million entrepreneurs who have small stores in Brazil that Inventa is targeting,” Salama said. “Our B2B marketplace connects brands and small retailers to help them with assortment in the areas of cosmetics, healthy food and home decor. They can also see what is trending, which makes recommendations more useful.”

The company offers over 7,000 products from 400 brands and has amassed more than 20,000 customers.

The new funding will enable Inventa to invest in its technology team — much of its 100-person workforce is in engineering — and to build its sales and marketing teams. Salama expects to grow massively in the employee area with a goal of adding another 400 people in the next few years.

He also plans to grow its brands to 10,000 as Inventa goes deeper into the cosmetics, healthy foods and home decor verticals. In addition, the company will focus on its technology development so that it can eventually offer a free software product for small suppliers and retailers.

“Amazon, MercadoLibre and Rappi are catering to the business-to-consumer world, but in B2B, there are much less companies targeting this market,” Salama said. “It is large, but there are no solutions, so we are ready to serve them.”

Spectrum Labs raises $32M for AI-based content moderation that monitors billions of conversations daily for toxicity

Two years into the pandemic, online conversations are for many of us still the primary interactions that we are having every day, and we are collectively having billions of them. But as many of us have discovered, not all of those are squeaky clean, positive experiences. Today, a startup called Spectrum Labs — which provides artificial intelligence technology to platform providers to detect and shut down toxic exchanges in real time (specifically, 20 milliseconds or less) — is announcing $32 million in funding. It plans to use the money to continue investing in its technology to double down on its growing consumer business and to forge ahead in a new area, providing services to enterprises for their internal and customer-facing conversations, providing not just a way to help detect when toxicity is creeping into exchanges, but to provide an audit trail for the activity for wider trust and safety tracking and initiatives.

“We aspire to be the leaders in language where civility matters,” CEO Justin Davis said in an interview.

The round is being led by Intel Capital, with Munich Re Ventures, Gaingels, OurCrowd, Harris Barton, and previous backers Wing Venture Capital, Greycroft, Ridge Ventures, Super{set}, and Global Founders Capital also participating. Greycroft led Spectrum’s previous round of $10 million in September 2020, and the company has now raised $46 million in total.

Davis, who co-founded the company with Josh Newman (the CTO), said Spectrum Labs is not disclosing valuation, but the company’s business size today speaks to how it’s been doing.

Spectrum Labs today works with just over 20 big platforms — they include social networking companies Pinterest and The Meet Group, dating site Grindr, Jimmy Wales’ entertainment wiki Fandom, Riot Games, and e-learning platform Udemy — which in turn have millions of customers sending billions of messages to each other day, either in open chat rooms or in more direct, private conversations.

Its technology is based around natural language and works in real time both on text-based interactions and audio interactions.

Davis notes that its audio work is “read” as audio, not transcribed to text first, which gives Spectrum’s customers a significant jump on responding to the activity, and counteracting what Davis referred to as “The Wild West nature of voice,” due to how slow responses typically are for those not using Spectrum’s technology: a platform has to wait for users to flag iffy content, then the platform has to find that audio in the transcriptions, and then it can take action — a process that could take days.

This is all the more important since voice-based services — with the rise not just of podcasting but services like Clubhouse and Spaces on Twitter — are growing in popularity.

Whether text or audio, Spectrum scans these exchanges for toxic content covering more than 40 behavior profiles that it built initially in consultation with researchers and academics around the world and continues to hone as it ingests more data from across the web. The profiles cover parameters like harassment, hate speech, violent extremism, scams, grooming, illegal solicitation and doxxing. It currently supports scanning in nearly 40 languages, Davis tells me, adding that it could work with any language, , although Davis tells me that there is no language limit.

“We can technically cover any language in a matter of weeks,” he said.

The most visible examples of online toxicity have been in the consumer sphere — where they have played out in open-forum and more private online bullying and hate speech and other illegal activity, an area where Spectrum Labs will continue to do work and invest in technology to detect ever more complicated and sophisticated approaches from bad actors. One focus for Spectrum Labs will be in working on ways to improve how customers themselves can also play a role in deciding what they do and definitely do not want to see, alongside controls and tools for a platform’s trust and safety team. This is a tricky area, and arguably one reason why toxicity has gotten out of hand is because traditionally platforms have wanted to take a hands-off, free speech approach and not meddle in content, since the other side of the coin is that they can also be accused of censoring, a debate that is still very much playing out today.

“There is a natural tension between what the policy implements and what users want and are willing to accept,” Davis said. His company’s view is that the job of a platform “is keeping the worst of the worst off, but also to provide consumer controls to make selection over what they want to see over time.”

Alongside that, Spectrum plans to move more into enterprise services.

The opportunity in enterprise is an interesting one, as it includes not just how people within a company converse with each other (which largely might take a similar form to the consumer-facing services that Spectrum Labs already provides), but also how a company interfaces with the outside world in areas like sales, customer service and marketing, and then leveraging the information that Spectrum Labs gathers in its analytics to potentially alter how each of those areas subsequently operates.

To be sure, this is not a market segment that has been ignored. Spectrum’s competitors here will include another startup in the conversation monitoring space, Aware, which focuses on enterprise exclusively. (L1ght, meanwhile, is another competitor in the consumer sphere.)

And there will certainly be others. We noted when we last wrote about Spectrum Labs that the founders and founding team came from Krux, a marketing technology company that was acquired by Salesforce (where they worked before leaving to found Spectrum Labs). I wouldn’t be surprised to see Salesforce taking a more interested role in this area in the future, not least because it is building out a very wide toolset to help companies run their businesses more efficiently, not just limited to CRM; but also because Bret Taylor, who once founded another social network and used to be the CTO of Facebook, is now helping to run the show, and may well have an especially informed grip on how communications forums can be used and abused.

For now, to address both the consumer and enterprise issues, Intel is coming in as a strategic investor in this round, Davis tells me. The plan will be to integrate Spectrum Labs’ technology to work more closely with Intel’s chip designs, which will increase the speed that it works even more, and Intel will be able to use as a unique selling point with Intel’s would-be hardware customers as they give a higher priority to trust and safety issues themselves.

“We believe Spectrum Labs’ Natural Language Understanding technology has the potential to become the core platform that powers the trust initiatives of thousands of companies around the world,” said Mark Rostick, VP and senior MD at Intel Capital, in a statement. “As digital trust and ethical operations emerge as a key factor to help organizations differentiate themselves, we see a huge opportunity to build a Trust & Safety tech layer into enterprise operations.”