Probably the worst interview experience I’ve ever had. Initial call with one of their member of technical staff and then they send you a take home assessment to build a Q&A agent based on the ConvFinQA financial dataset released in a 2023 paper. You are asked to use a different approach than the DSL approach used in the original paper. They are likely expecting an agentic approach. It is ridiculous that they don’t even provide a temporary API key for LLM costs but expect you to do extensive testing with frontier models and beat the score achieved in the paper to pass the interview. This can cost $10-$20 and they expect you to swallow it with no guarantee of even any feedback. They advise you to handwrite the final report with little to no AI assistance if you want to pass this assessment. Again a ridiculous idea given the assessment is already too much as it can take two full days of work even with AI. The worst part for me was that the agentic solution I built achieved a score well over that of the paper and I wasn’t even given any feedback on why my solution was rejected. They just seem to be establishing a database disguised as a technical assessment. Great way to get unpaid work done. They don’t seem to give a damn about candidates’ time and effort. Incredibly frustrating and I wanted future candidates to know about my experience.