Building a healthcare case management platform from the ground up with a small team. Hooked up the AI triage system using message queues and async task processing so clinical diagnosis models feed directly into the case workflow. Built the full lab ecosystem — registration, licensing checks, fee scheduling, per-payer insurance catalogs, and a bidding engine that scores labs on geolocation, network status, and other criteria then auto-awards the top pick. Developed AI-assisted ICD-10 to CPT mapping that filters labs by the procedures they actually perform and scores them into the bidding pipeline. Built the document and attachment layer handling test results, images, and medical records across cases and claims. Integrated FedEx API for medical kit shipping: label generation, tracking, and delivery updates.

Reduced microservice orchestration time from 2 seconds to 110ms by replacing chained HTTP calls in the SAGA architecture with gRPC. Built a real-time PII redaction system that runs live during screen recording without any LLM, using OCR, pattern matching, and intelligent frame analysis. Developed a workflow recording system that captures screen, audio, and user I/O, builds an action graph from multiple sessions, and replays tasks from natural language input — including the full recorder tool suite: replay engine, file management, system control, and I/O capture. Built field-level AES-256 encryption by overriding Django's ORM field classes, with RBAC-controlled decryption and deterministic encryption on searchable columns. Implemented microservice auth through Traefik with dynamic routing and RBAC. Developed a comprehensive AI evaluation pipeline tracking true/false positives, token usage, and accuracy across multiple dimensions.

Led development of an AI browser agent that executes multi-step browser tasks from natural language with no user intervention. Built the core execution engine reading multiple input signals in real time, with async background processing so the next action is ready before the current one finishes, and a per-action verification loop. Made the agent undetectable: randomized timing, natural cursor movement, stealth browser sessions. Took the product from concept to beta and owned the core agent logic. Co-developed the Chrome extension frontend and maintained backend services: auth, billing, usage metering, and AI model integration.

Built the backend from scratch for an AI-powered computer-use agent that executes tasks on behalf of users through natural language. Designed a microservices split with Node.js/Express handling auth, billing, and account management, and Django handling AI model integration and agent logic. Implemented OAuth-based multi-account linkage (Google, Zoho, and others) enabling the agent to call third-party APIs on the user's behalf. Designed data models across PostgreSQL and MongoDB for sessions, billing records, token usage, and agent activity logs. Integrated Stripe for billing and token metering; integrated OpenAI with prompt tuning, context management, response caching, and token streaming. Implemented rate limiting, CI/CD with Dockerized containers, and wrote unit and end-to-end test suites across services.

Built the backend for a real-time portfolio management platform serving ~700 users. Integrated Finnhub WebSocket streams for live financial data ingestion and developed portfolio tracking APIs for holdings, performance, and alerts. Implemented prediction features using technical indicators and in-house vector-based models. Worked across a microservices stack with Django, PostgreSQL, Redis, and Kafka, wrote unit test suites across services, and deployed production workloads on Google Cloud Platform.