Prototype AI Investigation Agent for Business & Document Evidence
Mosaic started as a document intelligence platform. It has now evolved into an investigation workflow that combines business analytics and document review in one application.
Enterprise investigations are rarely based on a single data source.
Business metrics live in databases, while supporting evidence lives inside documents. Switching between dashboards, SQL tools, and documents slows investigations and makes decisions harder to verify.
Mosaic explores how these workflows can be brought together in a transparent AI-assisted experience.
- Natural-language business questions
- Verified SQL generation
- DuckDB execution
- Evidence ranking
- Business recommendations
Proof
- Receipt search
- Missing-field detection
- Review queue
- Human review workflow
Proof
Python • Streamlit • DuckDB • Pandas • Plotly • FAISS • Sentence Transformers
Datasets: Olist • SROIE
✅ Working prototype
Current focus:
- Transparent investigations
- SQL-backed evidence
- Human-in-the-loop review
- Explainable recommendations
pip install -r requirements.txt
python src/olist_loader.py
python src/sroie_loader.py
streamlit run src/app.py