Senior Software Engineer — Factiiv.io
- Architected and shipped a multi-tenant AI agent-builder SaaS platform, owning the full stack from Node.js/TypeScript REST APIs and PostgreSQL schema design through to the React front end; currently serves 1,000+ API requests daily.
- Built production LLM features on OpenAI and Claude with custom RAG pipelines over Qdrant vector DB (LangChain, LlamaIndex), including chunking, embedding and re-ranking strategies for multi-turn, context-aware responses.
- Cut hallucinations by ~90% via multi-chain prompting, tool/function calling and optional live web-search retrieval; fine-tuned a GPT model for domain summarisation, improving coherence 15%.
- Integrated multiple third-party APIs into the platform — LiveKit plus Google Cloud STT/TTS, VAD and turn detection — to deliver real-time bi-directional voice agents with sub-second latency.
- Built a self-learning LLM feedback loop that captures conversation outcomes and user signals from both the chat and voice interfaces, feeding them back into retrieval ranking, prompt templates and fine-tuning datasets so agent responses measurably improve with usage — lifting resolution quality and customer experience across both channels.
- Orchestrated asynchronous background workers for model-serving, ingestion and vector-store sync, keeping long-running AI jobs off the request path.
- Ran code reviews, set backend/AI architecture standards, and mentored junior engineers; continually evaluates emerging frameworks (LangGraph, RAG Fusion, agentic architectures) and brings the useful ones into production.