Agentic Medical Claim Analysis
An open-source Streamlit app running a panel of configurable AI agents over medical claims spreadsheets to analyze diagnosis consistency, upcoding, and service frequency.
Open source · Live apps
Tools from real engagements, published openly.
Working Streamlit and Python applications across insurance, healthcare, finance, energy and applied AI. Each one started as a tool built for client work. Open them, try them on your own data, and read the code.
- Live apps
- 9
- Domains
- 6
- Cost to use
- Free
- Stack
- Python
The collection
Pick a domain.
Apps may be asleep. They run on Streamlit’s free tier and idle after a while. If one shows a sleep screen, press “Wake up” and give it a few moments.
Global Markets Report
Interactive dashboard for getting a high level overview of global financial markets.
RUL Analyzer
Interactive application allowing users to simulate remaining useful life for a component.
Reinsurance Pricer
High-performance Monte Carlo excess-of-loss (XoL) pricing engine designed for reinsurance actuaries to evaluate layer losses, tail volatility and technical premium.
Medical LOB Reserving Tool
An actuarial reserving workbench for a medical line of business. Upload a claims workbook to build the development triangles, then compare Mack chain ladder, bootstrap ODP, Cape Cod, Bornhuetter-Ferguson and credibility-blended reserve estimates, with a recast back-test and a seasonality check.
NRC Technical Letter Search
Grounded question answering over NRC technical letter reports. Every statement carries a verbatim quote checked against the source page and highlighted there; anything that fails verification is withheld rather than shown.
PDF → Structured CSV
A Streamlit app that turns unstructured PDFs into one tidy CSV using Gemini's native document understanding. The model infers a schema from a sample document, allowing you to refine it before bulk extraction.
Ask Your Data
A Streamlit app showing how large language models can work with structured data, not just text. Ask a business question in plain English and an SQL agent writes the query, runs it read-only in DuckDB over your CSVs and fixes it if it fails. An analyst agent writes up the answer, a visualization agent picks the chart, and every answer shows its SQL, raw results and agent trace.
FraudLens: Transaction Fraud Detection
A Streamlit demonstration of real-time card fraud detection for a bank, built on a synthetic transaction world with six fraud typologies. Every score comes with SHAP contributions, Bayesian posteriors and a complete, downloadable audit trace.
Built with
- Python
- Streamlit
- AI agents
- RAG
- Gemini
- HuggingFace
- spaCy
- XGBoost
- SHAP
- Bayesian inference
- Monte Carlo
- DuckDB
Want one of these on your own data?
Every app here is a smaller cut of something we deploy for clients. We can adapt any of them to your systems, your data and your controls.