Now: Building agentic AI workflows for clinical data standards at ZS.
> ls projects/
End-to-end work, from framing the problem to deployment. Each case study shows which stages I owned.
Point at a star to see its pattern; open it for the case study. The patterns are decoration, not results.
Work done in a professional role. Source code and internal systems are not public; each case study covers the problem, my role and the results I can share.
An event-driven, multi-agent AI platform. It spots manufacturing orders at risk of missing their delivery date, finds the root cause from evidence, plans a recovery, and carries it out after a project manager approves.
AI suggestions for clinical SDTM mapping. Agents validate and generate annotations and mapping details, and data programmers review, edit and accept them, with full audit history.
An error-resolution agent for R code generated by the ROAK framework. It matches runtime errors to a curated rulebook, explains the cause and returns a known fix, so programmers don't have to wait for the core team.
LLM agents read thousands of unstructured manufacturing documents and build a knowledge graph of suppliers, customers, materials and capabilities. The graph powers supplier matching, supply-chain risk detection and cross-business-unit discovery.
An agentic workflow that independently re-creates the analysis results behind clinical tables, listings and graphs (TLGs), then reconciles them with the production outputs: an automated second check.
An LLM copilot that turns a procurement request into a vendor recommendation report. It reads the request, finds and ranks vendors with semantic search, checks compliance, and writes up the recommendation.
A multi-modal pipeline that turns hours of supplier audit recordings into a small set of distinct, searchable keyframes with defect summaries.
Forecasts EV battery State of Health (SoH) up to 12 months ahead from high-frequency telemetry, with 200+ engineered ageing features and explainable gradient-boosting models.
Independent projects, with links to code where available.
Behavioral cloning: a CNN based on NVIDIA's end-to-end architecture learns to steer a car from recorded driving in a simulator.
Draw a digit by hand; a CNN recognizes it and grows your virtual plant.