I'm an AI engineer currently building SaaS products in production. My core craft is AI — RAG systems, AI agents, and production chatbots — integrated into existing products or built from scratch. Web and IoT are the surrounding skills that let me ship end to end.
Multi-tenant donation management platform for charitable fundraising. Organizations onboard, run campaigns (incl. micro-sponsoring and Advent Calendar drives) and manage donor relationships; donors discover projects and track their giving. Stripe-powered payments and subscription tiers, role-based access, GDPR-compliant, Dockerized deployment.
Full-stack platform for managing and automating software testing workflows on an enterprise validation framework. Test execution is dispatched to remote agents through a multi-component orchestrator (Receptionist · Observer · Event Bus · Killswitch). Pluggable GitHub / GitLab integration, environment health checks, fine-grained RBAC.
Full-stack web app that verifies user-submitted claims with an evidence-backed verdict, citations, and confidence score. React + Tailwind frontend, Node/Express REST API (JWT, MongoDB), and a Python LangChain / LangGraph agent that decomposes the claim, runs DuckDuckGo search, scrapes top sources, cross-checks Google's Fact Check Tools API, and uses GPT-4o (OpenAI / OpenRouter, pluggable) to produce a strict Pydantic-typed JSON verdict. Dockerized with a token-cost calculator for predictable LLM spend.
Internal knowledge assistant for a SaaS support team. Indexed ~8k help-center articles, tickets and runbooks with hybrid retrieval (BM25 + pgvector) and a cross-encoder reranker, grounded answers with inline citations. Evaluated with Ragas — 92% faithfulness, retrieval recall@5 of 0.87 — and wired to Langfuse for online monitoring.
Agent that drives a real browser to fill forms, scrape dashboards, and search the open web on behalf of the user. Plans a multi-step task, calls tools (Playwright, search, scraper), recovers from failures, and returns a structured result. Built on the same tool-use patterns as my fact-checking platform — full observability, retry logic, typed outputs.
Customer-facing chatbot integrated with Zendesk for a B2B SaaS. Streams answers grounded in the product knowledge base, handles auth-gated account questions through tool calls, and escalates cleanly to a human agent with full context handover. Deflected ~38% of tier-1 tickets in the first 60 days, with a 4.4/5 CSAT on resolved chats.
Wake-word voice assistant running on an ESP32 with an I²S mic and speaker. Audio is streamed over MQTT to a small gateway that handles VAD, Whisper transcription and an LLM call, then synthesizes a TTS reply back to the device. End-to-end latency under 2s on local Wi-Fi. The same pipeline doubles as a sensor → LLM-summary loop for a connected weather station.
Two years in, AI is where I'm strongest. I design and ship RAG pipelines, build multi-step AI agents with tool use, and deliver chatbots that hold up in production — with evals, guardrails, and observability behind them. In parallel I ship enterprise full-stack work on Java + Angular SaaS platforms (donations, test automation), so I can carry a feature from prompt to UI to deployed system end to end.
Two years building production software end to end — currently focused on AI: RAG systems, agents and chatbots integrated into SaaS products. Comfortable across the full stack with AI as the core craft.
Freelance work on connected hardware and web platforms. Owned the full stack across multiple products — ESP32 firmware, MQTT pipelines, backend APIs and React dashboards. Delivered, deployed, in real use.