Answers from SEC filings, each one traceable to its source
Our founder built the retrieval, answering and evaluation pipeline.
Context
Analysts spent hours reading 10-K and 10-Q filings and earnings-call transcripts to answer routine questions and write summaries.
The problem
A general chatbot answers fluently and cites nothing, which is useless when an analyst has to stand behind every number.
What was built
- Retrieval-augmented generation over filings and transcripts, using FAISS, SentenceTransformers and GPT-4
- Question answering where each answer is traceable to the passage it came from
- Generated summaries scored against reference summaries with BLEU, so quality was measured rather than assumed
The client is not named because the work was done under a confidentiality agreement.
More work
alphashots.ai, a retail trading analytics product
Analytics for 20,000 retail users at a tenth of the usual cloud cost
A US K-12 school network of about 50 campuses
A quality gate for AI-generated content: 50,000 items checked, about 1 in 10 caught
Open source, built by our founder
UnvibeCode: what AI-written code actually does
Tell us what you have and what is going wrong.
A 20-minute call, then a written scope with acceptance criteria and a fixed price within one business day. If it isn't a fit, we'll say so on the call.
Or email hello@destello.us or call (628) 202-5322. We reply within one business day.