I build AI systems that run in production. The last year and a half went into an LLM-driven anomaly detection pipeline on live crypto market data (Anthropic Claude API, Go, NATS JetStream, TimescaleDB) that replaced manual review, and my own AI job-application platform with multi-provider model routing and pgvector semantic caching. Underneath that sit 10+ years of backend engineering: event-driven microservices, real-time systems, multi-tenant SaaS.

Day to day I work across LLM applications, agentic and tool-calling systems, RAG, fine-tuning with LoRA/QLoRA, and the MLOps that keeps models serving.

LLM & Agents

RAG & Vector Search

ML & Fine-Tuning

MLOps & Cloud

Data & Backend

Daily tooling

Anthropic Claude API · Go · NATS JetStream · Redis · TimescaleDB · WebSockets · Docker

LLM-driven pipeline that detects anomalies and patterns in real-time crypto market data and replaced manual review. It runs on a 4-service event-driven platform: an 8-stream NATS JetStream topology, a pub/sub WebSocket layer for concurrent clients, Redis split across 4 databases (caching, aggregation state, rate limiting, connection tracking), and TimescaleDB for time-series persistence.

FastAPI · HTMX · Postgres 16 + pgvector · Redis · Docker · Chrome extension

Ingests job postings and drafts tailored resumes and screening answers, every claim traced back to a structured profile; a human reviews and submits each application. LLM calls go through OpenRouter across three model tiers with automatic fallback and hard daily and monthly budget caps. A 3-layer cache (exact match, pgvector embeddings, provider prompt cache) keeps cost down. The Chrome extension captures postings and pre-fills forms; login and final submission stay manual by design.

• Subke CRM & Logistics Platform: multi-tenant enterprise CRM with database-driven runtime RBAC, real-time messaging hub, QR-code parcel tracking and JTL ERP integration. 30+ domain service classes across 72+ migrations, delivered with AI-assisted development alongside hand-written code.
• websocket-server-market-streamer: Node.js/TypeScript relay that subscribes once to upstream Binance streams and re-broadcasts normalised market data to downstream clients, with Redis catch-up caching.
• Bryant Dental: e-commerce platform for a UK dental equipment manufacturer. Awwwards Honorable Mention 2018, Dental Industry Awards Best Website 2019.