AI & Computer Vision Engineer specializing in agentic AI systems, LLM evaluation & observability, and edge-optimized computer vision.

AI & LLMs:

Vision & Edge:

Infra & Cloud:

• Custom ReAct Agent Framework — Framework-free agent loop with Pydantic validation, exponential backoff retries, and self-correction. Designed for reliability in high-stakes domains.
• LLM Fine-Tuning Pipeline — Domain-adapted 7B–16B models (StarCoder family) using DAPT + LoRA, significantly reducing generation latency and improving task-specific quality.
• End-to-End Observability & Evaluation — LangFuse tracing + RAGAS/FastAPI evaluation service achieving strong attribute-level performance across complex outputs.
• Edge Vision Systems — Optimized vision models deployed via ONNX/TFLite with consistent sub-20ms inference latency.

Production-oriented medical reasoning system featuring:

• Asynchronous ReAct agent with real-time SSE token streaming
• Redis Pub/Sub + Celery distributed workers
• LLM-as-a-Judge evaluation + full LangFuse tracing
• Concurrent-safe SQLite (WAL mode) persistence
• Rate limiting, self-healing recovery, and structured audit logging

Architecture prioritizes reliability, observability, and low-latency user experience under concurrent load.

• Agentic AI Systems — Comprehensive notes on ReAct, Reflection, Code-as-Action, Tool Sandboxing, Tracing, Evals, and Multi-Agent patterns
• Additional computer vision and deep learning implementation notes available in repositories