A research assistant that runs on your own computer
Reads an Excel sheet, answers, and checks its own answers before accepting them.
A working sample we built to show how we would solve a problem like this. It is not a delivered project.
In plain terms
A research assistant that runs entirely on your own computer. It reads an Excel sheet, looks up relevant information, writes an answer, and then checks its own answer for relevance and support before accepting it. When it is not confident, it says so instead of guessing.
- Runs fully on the user's computer, so no data leaves it.
- Checks its own answers and retries a limited number of times.
- Works from a spreadsheet, so no technical skills are needed.
What this does not claim: Its search is basic (one method, no re-ranking) and it has no formal accuracy test, so it should not be taken as a complete search-evaluation project.
Technical details
Self-RAG-style loop: grade retrieval relevance, generate, grade groundedness and usefulness, then retry or regenerate up to a limit, otherwise report low confidence. Strict JSON-schema grading prompts; dense retrieval only; Excel I/O so a non-technical user can drive it from a spreadsheet.
- Python
- Chroma vector store
- Ollama (local embeddings)
- pandas / openpyxl
Need something like this?
Tell us the problem in your own words. We reply with a plain, practical plan.
Discuss a project like this