First Chair is built and run by Nikhilvarma Kandula, a founder and AI engineer in Germany. He spent eighteen months in US fintech before this. One person writes the scan engine, reads the answers, and answers the support email.
The best data work is invisible. The pipeline nobody notices because it never breaks.
Eighteen months at MicroIntech, a US fintech, went from data engineer to lead developer. The work was rebuilding a monolithic financial platform into event-driven microservices carrying 500+ concurrent users at full transaction integrity, and wiring an LLM audit pipeline that cut fifteen hours of weekly manual review down to about three.
The habits behind this product come from there: idempotent decisions, logged propensities, and the assumption that anything unlogged did not happen. It is also where the confidence routing came from: the engineering that sends a doubtful case to a human instead of letting a model guess.
Alongside it, Lead Developer at EngineeredPrompts, a premium AI-prompt platform. He built the prompt library and model orchestration behind it, the paid tier end to end, and led the two developers on it. That title is published on their own team page, which is the only version of the claim worth anything.
Now in Germany reading for an M.Sc. in Big Data & Business Analytics at FOM Hochschule through August 2027, alongside peer-reviewed research on rainfall estimation by data fusion. It reached a probability of detection of 0.58, beating Kriging with External Drift on the same dataset.
18
Months in US fintech
13
Projects written up
4
Products live in production
1
Peer-reviewed paper
Not values. Four positions paid for at least once, each one visible in a decision this product actually made.
A percentage quoted afterwards is unfalsifiable. So the visibility score prints its own arithmetic next to the number: recommended 1.0, first-mentioned 0.6, mentioned 0.4. A partner can check it.
In a financial audit a confidently wrong answer is worse than no answer. The same holds here: the engines are non-deterministic, so First Chair stores every response verbatim and never summarizes away the text a score came from.
A portfolio of only confirmed hypotheses is a portfolio that has quietly deleted its failures. If a scan finds nothing moved, the report says nothing moved.
If the product cannot be explained in a paragraph a non-technical reader follows start to finish, the problem is not yet understood. That paragraph is the first thing on the home page.
Each of these is running, not a screenshot in a deck. First Chair is the one you are reading.
At counsel table, first chair is the lawyer who runs the case. That lawyer stands up, examines the witness, and answers for what ends up in the record. Second chair does real work and is not the name anyone repeats afterwards.
An AI answer is a record too. Asked for the best firm in a city, an engine names two or three, and the rest of the market is not in the transcript at all. This product exists to tell you which chair you are sitting in today, and to keep the receipt.
The mark is the same idea reduced: a counsel table with three seats, the first one taken.
Questions about the scan engine, the evidence behind a number, or working together. Email reaches a person, usually the same day.
A free audit runs twenty questions against ChatGPT, Gemini and Perplexity and returns the answers verbatim.