AI/ML Engineer focused on agentic systems, LLM orchestration, and applied computer vision — with full-stack engineering as the delivery layer that turns those systems into shipped products.

Most of my recent work sits in one lane: designing agents and models that make decisions (task orchestration, safety review, detection), then wrapping them in a production-grade interface (FastAPI/WebSocket backends, React/Next.js frontends) so they're actually usable. The full-stack projects below are the same skill set applied to standard SaaS/business platforms — proof that the architecture and delivery discipline carries over regardless of the problem.

Stack, chosen by project fit, not everything at once:

• AI/ML core: Python, FastAPI, WebSocket streaming, local LLMs via Ollama, OpenAI/Gemini APIs, scikit-learn, OpenCV, YOLOv8/YOLOv11
• Delivery layer: React 19, Next.js, TanStack Start (SSR), TypeScript, Node.js/Express, Tailwind CSS
• Data & infra: PostgreSQL (with Row-Level Security), MongoDB, Supabase, Docker, Cloudflare Workers, microservice/gateway architecture

• Agentic AI — multi-agent orchestration, permission gating, tool-use verification
• LLM safety & guardrails — review pipelines that sit between an agent and its execution environment
• Applied computer vision — object and gesture detection with the YOLO family
• Full-stack product delivery — turning the above into usable, deployed systems (RBAC, SSR, RLS, REST APIs)

- Multi-agent orchestration & permission gating
- Model Context Protocol (MCP) integration
- RBAC architecture
- Row-Level Security (RLS)
- Server Side Rendering (SSR)
- REST API design
- Authentication & authorization
- CI/CD fundamentals

Primary focus. Ordered by how recent and representative each project is.

lato-validation — local-first multi-agent task orchestrator with real-time WebSocket token streaming, permission gating, and empirical action verification. Built on FastAPI + Ollama for local LLM inference, with a ReactFlow front end for visualizing agent execution. Most recently active project. agentic-ai multi-agent local-llm fastapi ollama websocket reactflow python typescript
sentinel-mcp — a three-stage safety guardrail agent for LLM coding assistants (Claude Desktop, Cursor, CodeX), shipped as a plug-and-play MCP plugin. Pipeline: rules engine, trained TF-IDF/logistic-regression classifier (76% CV accuracy), then LLM review for ambiguous cases. Supports both stdio and SSE transport. Live demo python security claude fastapi guardrails ai-agent llm ollama mcp-server
Development-of-a-Multi-Turn-Conversational-AI-Bot — GPT-4-powered customer service bot that handles multi-turn conversation state for a real product (Split Money app support).
voice-notes-ai — AI-driven voice capture and processing pipeline in Python.

Supporting skill set — the same engineering discipline (architecture, RBAC, SSR, clean APIs) applied to standard business platforms, proving it isn't tied to one narrow use case.