Builder. Engineer. Product thinker.
I build systems end-to-end across AI, fintech, healthcare and distributed platforms — from architecture and APIs down to data models, failure handling, tests and production behavior.
I use coding agents aggressively because they increase implementation throughput, but I still own the architecture, decomposition, code review, debugging, validation and system behavior.
I create a lot of proofs of concept projects here to explore many different cutting edge paradigms(curently 45+ such projects)
- SiteReliability — Multi-tenant observability platform with FastAPI/SQLAlchemy, adaptive EWMA anomaly detection, async ingestion connectors, webhook delivery, JWT/API-key security, tenant isolation, and a 266-test validation suite.
- Mailchimp Trends Engine — Production-shaped signal intelligence pipeline spanning ingestion, normalization/dedup, clustering, trend detection, grounded RAG, personalization, campaign workflows, lineage, retries/circuit breakers, and local-to-AWS adapter composition.
- Adaptive Resource Router — Capacity-aware request-routing POC with pluggable eligibility constraints and selection strategies for distributing work across heterogeneous resources.
- BiotechOS — Large closed-loop operating platform spanning predictive AI, workflow orchestration, async execution, resource routing, multi-tenant controls, typed state machines, external-system integration, and robotic wet-lab automation.
- Ops Automation — Workflow-automation platform POC focused on durable process execution, task orchestration, state transitions, retries, and operational visibility.
- Voice AI Revenue Call Intelligence — AI revenue-operations platform that converts unstructured calls into structured deal stages, risks, stakeholders, next steps, and operational dashboards.
- Grounded RAG Support Agent — Grounded support-agent POC combining retrieval, tool execution, evidence-constrained answers, and evaluation of agent behavior.
- Clinical Policy Review Agent — Healthcare decision-support prototype spanning prior-auth intake, OCR, de-identification, policy-grounded citations, structured review, and human-in-the-loop nurse decision support.
- Retail AI/LLM based Analytics — Conversational analytics POC combining NL-to-SQL, semantic business rules, query validation, and grounded natural-language answers over structured data.
- Context & Privacy-Aware Interactive Outreach — Stateful multi-channel agent workflow with context resolution, persistent sessions, approved-content retrieval, privacy-state enforcement, verified-intent transitions, and auditable interaction history.
- Insurance Intake Intelligence — Document-intelligence architecture slice using schema-guided LLM extraction, evidence grounding, confidence scoring, reviewer correction, and quality/operations metrics.
- GrantAssistant — Chrome-extension + Next.js AI copilot integrating Claude, Supabase OAuth/Postgres, RLS, citation ranking, constrained text transformation, and an auditable PDF compliance trail.
- CRO AI Workflow — Sanitized architecture of a commercial LLM workflow using structured intake, typed normalization, independently recoverable generation stages, human review, versioned persistence, and operational evals.
- BiotechOS RFP Tool — Applied-AI workflow prototype around biotech/CRO request processing and structured response generation.
- Secure File System Storage POC — Tokenizing storage-wrapper POC using envelope-style per-store keys to separate logical identifiers from protected underlying storage.
- Multi-tenant healthcare API — Multi-tenant healthcare API POC demonstrating tenant isolation, PHI redaction, policy-aware access, and auditable request trails.
- ECG-Based Authentication — Physiological-identity prototype spanning ECG preprocessing, fiducial/non-fiducial feature extraction, Random Forest inference, waveform analysis, and a runnable authentication workflow.
- Live-Presence Authorization — Security architecture prototype combining physiological liveness signals, deterministic asset policies, signed JWT authorization assertions, replay/spoof checks, and tamper-evident audit logging.
- Tokenized Reserve → Treasuries — Settlement prototype connecting Solidity/Hardhat token rails, merchant payment flows, reserve redemption, Next.js UX, and server-side brokerage/Treasury execution adapters.
- Brain Stroke Detection — YOLOv8 medical-imaging inference prototype with Streamlit, Firebase authentication, Google Drive integration, and an end-to-end image-screening flow.
- Onco VisionAI — Multi-model cancer-imaging prototype combining MRI, dermoscopy, histopathology, ultrasound, mammography, and blood-cell inference behind a common application interface.
- Retina AI Detection — Retinal screening pipeline combining YOLO lesion localization with ONNX disease classification, DR grading, vision-risk classification, and local/cloud execution modes.
- Brain-Computer Interface — Real-time EEG/P300 system spanning signal acquisition, preprocessing, multiple deep-learning architectures, LSL streaming, Flask inference, hardware experimentation, and assistive-device interaction.
- Architecture through implementation
- Strong contracts and clear system boundaries
- Boring infrastructure unless complexity earns its place
- Tests, evals and observability are part of the build
- AI-native development with Claude Code, Codex, Cursor and agents
- Prefer working systems over diagrams that have never been exercised
20+ years across software, AI, healthcare, fintech and large-scale systems. Engineering experience across Philips, Toshiba, Hitachi, GE and Microsoft research. MS CS — USC.
Based in the Bay Area.