Pandas wants to give Latin American businesses buying power in Asia

Access to global supply chains can be difficult for small businesses in Latin America, but companies like Meru, which raised funding in March to source and import goods between Mexico and China, and now more recently Pandas, are tapping into overseas relationships and technology to make this easier.

In Pandas’ case, the company is doing something similar to Meru, but starting in Colombia, connecting small businesses directly with Asian manufacturers, so that they can reduce the high fees often imposed by half a dozen importers and intermediaries as well as logistical problems that all businesses are facing right now where inventory is now taking many more months to arrive than during pre-pandemic times.

Co-founders Rio Xin and Marcos Esterli started Pandas just three months ago to provide Asian-origin inventory to micro-businesses in Latin America. Their collective background includes careers at McKinsey and Treinta for Esterli, and McKinsey, with more than seven years spent in China, for Xin, where he told TechCrunch he developed a strong network in the region.

“The main issue that we’ve seen is people who don’t understand the Chinese language or how Chinese manufacturers work and then you add in the logistical problems,” Xin added. “We are able to bridge the breach, while at the same time having our team in China to overcome all these logistics problems.”

Pandas B2B marketplace

Pandas B2B marketplace. Image Credits: Pandas

Here’s how it works: Businesses order products via the Pandas marketplace, touting lower pricing, in which the business can make purchases in a few clicks. Pandas takes it from there, offering one-day-delivery and customer support.

Esterli explained that people in Latin America have been using smartphones for their personal finances and other tasks, but that has not translated as quickly to the business side.

“A lot of customers told us Alibaba was something they wanted to use, but that it was very complicated to figure out,” he added. “We wanted to build an easy solution that was super intuitive because business owners don’t have that time to spend.”

Initially providing basic electronics products — think headphones, accessories and cables — and with a new round of funding, $5.8 million pre-seed, Pandas will move into categories like textiles and home accessories. The company touts the pre-seed investment as “the largest pre-seed financial in Spanish-speaking LatAm to date.”

Third Kind Venture Capital led the round and was joined by Acequia Capital, Picus Capital, Tekton Ventures, Partech, Liquid2 Ventures, Clocktower Technology Ventures, Gaingels and a host of individual investors, including Tul’s Juan Carlos Narvaez, Jose Jair Bonilla from Chiper, Treinta’s Man Hei and Lluís Cañadell, Pablo Viguera from Belvo, Nowports’ Alfonso de los Rios, Sujay Tyle from Merama and Ironhack’s Gonzalo Manrique.

So far in its young journey, the company is growing 100% month over month and has amassed a supplier network of about 300 out of 5,000 in China, Xin said.

In addition to moving into those new inventory categories, the new capital will enable Pandas to scale its operations, technology and product development and make new hires.

Xin expects to be in most of the main markets across Latin America in the next three years. In the meantime, new features coming down the pipeline in the next 12 months include a suite of fintech and analytics tools like financing.

Deepset raises $14M to help companies build NLP apps

Natural language processing (NLP), the field of AI that involves parsing text for tasks including summarization and generation, is a fast-growing technology. According to a 2021 survey from John Snow Labs and Gradient Flow, 60% of tech leaders indicated that their NLP budgets grew by at least 10% compared to 2020, while a third said that their spending climbed by more than 30%. Fortune Business Insights pegged the NLP market at $16.53 billion in 2020.

Against this backdrop, Deepset, the startup behind the open source NLP framework Haystack, today announced that it raised $14 million in a Series A investment led by GV with participation from Harpoon Ventures, System.One, Lunar Ventures, and Acequia Capital. The capital infusion arrived alongside Deepset Cloud, a new subscription product for building NLP-powered software.

“Driven by [our] belief in open source, the Deepset team has … been contributing models and research outcomes to the open source NLP community [for years],” Rusic told TechCrunch via email. “Haystack, the company’s flagship open source product, was born out of the experiences, expertise, and know-how gained while building NLP for large organizations and the need for a proper set of building blocks for scalable, API-driven NLP backend applications.”

CEO Milos Rusic cofounded Deepset with Malte Pietsch and Timo Möller in 2018. Pietsch and Möller — who have data science backgrounds — came from Plista, an adtech startup, where they worked on products including an AI-powered ad creation tool.

Haystack lets developers build pipelines for NLP use cases. Originally created for search applications, the framework can power engines that answer specific questions (e.g. “Why are startups moving to Berlin?”) or sift through documents.

Haystack can also field “knowledge-based” searches that look for granular information on websites with a lot of data or internal wikis. Rusic says that Haystack has been used to automate risk management workflows at financial services companies, returning results for queries like “What is the business outlook?” and “How did revenues evolve in the past years?” Other organizations, like Alcatel-Lucent Enterprise, have leveraged Haystack to launch virtual assistants that recommend documents to field technicians.


A screenshot of the Haystack interface.

According to Rusic, the goal with Haystack was to enable developers and product divisions to build modern, API-driven NLP apps successfully — and quickly. He notes that, while it’s often straightforward for a data science team to come up with a prototype, challenges can arise in transitioning from prototype to production. About 80% of AI projects — including NLP projects — never make it into production, according to a 2019 Gartner survey.

“[With Haystack,] development teams … are equipped with all the components to build a full-stack NLP application and are guided with the proper workflows … Modern NLP moves very fast, and it’s much easier to bridge the gap between the cutting-edge research and the actual production-ready technologies through open source,” Rusic said. “[Prebuilt NLP systems] are the basis [for Haystack] and often provide great results in pipelines without additional training. Customization, if needed, happens with end users and experts who provide feedback by testing and using new iterations of a [system] or a pipeline.”

But not every company chooses — or wishes — to go the DIY route. For those preferring a managed solution, there’s the aforementioned Deepset Cloud, which supports customers across the NLP service lifecycle. The service starts with experimentation — i.e., testing and evaluating an app, and adjusting it to a use case, and building a proof of concept — and ends with labeling and monitoring the app in production.

“All NLP services that are developed [with Deepset Cloud] can be used in any end application, simply by integrating an API,” Rusic said. “Example applications are NLP-driven enterprise search (think ‘modern Google-like’ search) and knowledge management.”

With the new financing secured ($15.6 million in total), Deepset aims to translate its open source success — thousands of organizations currently use Haystack — into increased revenue. Rusic says that the 30-person, Berlin, Germany-based company was bootstrapped and break-even before raising its first funding round in 2021, and now has large enterprise customers including Airbus.

“[With the new funding,] we’ll continue to build the open source Haystack NLP project — adding more features, making it even more straightforward for NLP-savvy backend developers to create NLP services,” Rusic said. “[We’ll also] develop Deepset Cloud into a fully-fledged enterprise software-as-a-service to build language-aware applications. This will include enabling more flexible workflows, more granular product lifecycle guidance, and offering essential and supplemental tools, like labeling and data integrations.”