For this post in the #LaunchWithAI series, I met with Azzeddine Chaibrassou, founder of Qard, a fintech startup, based in Paris, France, innovating in credit risk management. Qard started by addressing a niche pain point felt by many in the industry and took it to the next level moving swiftly from idea to MVP, all while launching the startup with their core in AI. Read on to find out what we discovered.
How Qard got started
“The main idea behind Qard was to give access to reliable information about companies to fintech, brokers, venture loan companies, neobanks, and other financial institutions. Today, disclosure of financial information about companies is registered as a legal act in France and other countries around the world. Yet, not many products existed that would facilitate bringing this information meaningfully to the consumers, to make the data actionable. Qard solves this problem by facilitating connections to various data sources, collecting relevant information, and bringing them together, either through an API or a dashboard.”
Moving from idea to MVP
“Going from idea to execution and MVP is one of the most resource and time intensive phases for a startup. Technologically, our product goal was to source and scan millions of documents, detect patterns and signals about the financial health of companies from the data set and eventually relay these patterns to our customers with our APIs. From the get-go, we’d have needed recruitments efforts to get the right DevOps and data engineering pipeline, and the in-house development would’ve taken months.
Having led teams through such 0 to 1 product creation, our head of engineering, with more than 15 years of industry experience, directed us towards leveraging high-performance Azure models to accelerate our time to development. So, we adopted and used Azure’s tools, which saved us a lot of time. Having tools dedicated to what you want to achieve saves time in technical development, but also in product management. Using ready-to-use solution Azure APIs vs installing and maintaining in-house ML pipelines freed the time for our data scientists to manage more complicated projects.”
Leveraging NLP for the Qard solution
“Qard uses natural language processing (NLP) to extract and explain useful information from the millions of documents available. From this data, we detect patterns and signals about the financial health of companies and relay this information to our customers.
This process, overall, has the following major components:
- Industrialized document processing with Azure OCR services and Azure Databricks. Databricks helps us by making the configuration of a multiprocessor’s cluster easier. We configure a cluster with CPU-oriented machines to process in tandem with the source’s files.
- Named Entity Recognition (NER) on the output of OCR, giving richer and more accurate information on the retrieved signal. Example: Extracting the entities of interest (e.g., organizations, people, places), linking these entities to corresponding concepts (e.g., type and name of the identified organizations) and providing enriched information to our customer through APIs hosted on Azure VMs.
Through this process, we’ve already served tens of thousands of PDFs.”
Taking your MVP to market
“Our API now provides information and data processing like never before. Our technical team still consistently faces optimization challenges. Our Microsoft mentor has been helping us work through these challenges with which we continuously improve and A/B test to establish our product-market fit.”