Projects

LLM systems · safety-critical routing

A multi-stage AI triage pipeline built on LLM structured outputs — classifying incidents, generating de-duplicated follow-up questions, and synthesizing incident records — with deterministic fallback logic for graceful degradation when models fail. A safety layer independent of the LLM force-escalates life-critical hazards, and a six-state routing-readiness engine gates automated dispatch on AI confidence, evidence completeness, cost tier, and tenant responsibility.

• Python
• FastAPI
• LLM APIs
• Structured Outputs
• LangChain

Backend architecture · billing & auth

A modular FastAPI backend spanning multiple services for investment analysis, portfolio management, billing, and mobile APIs. Includes Stripe-based subscription and entitlement infrastructure with webhook-driven billing updates and idempotency safeguards, hardened Google OAuth with resource-ownership authorization, and core workflows for deal versioning, stress testing, and actual-vs-projected performance evaluation.

• Python
• FastAPI
• PostgreSQL
• Stripe
• Google OAuth
• React Query
• Docker

ML systems · model serving

Translates survival-analysis research (MTLR, individualized survival distributions, conformal prediction) into production ML services. End-to-end pipelines cover preprocessing, model training, calibration, and evaluation, served through a scalable model-serving API that decouples real-time probabilistic inference from application logic.

• Python
• PyTorch
• FastAPI
• MLflow
• AWS
• Docker

Research concept · human-computer interaction

A long-term research and product concept exploring neural intent decoding and intelligent human-computer interaction. Currently at an early exploration stage — direction and scope are still being defined.

• Deep learning
• Signal modeling
• HCI