JI is Jason’s AI-powered assistant. It blends multiple layers of technology:
- Knowledge + Retrieval: A local knowledge base (Jason’s experience, résumé data, and curated context) plus a lightweight retrieval layer ensures answers stay grounded in verified facts.
- Reasoning & Language: Powered by GPT-5, the large language model running the conversation, which interprets questions, stitches retrieved context, and generates precise, natural answers.
- Safety & Guardrails: A context-first pipeline — if the data isn’t present, JI simply says it doesn’t know, avoiding hallucination.
- Voice Output: ElevenLabs neural TTS for lifelike voices, with automatic failover to Azure Cognitive Services if needed.
- Backend Stack: FastAPI + Uvicorn backend with async endpoints, serving both web UI and API; Bootstrap/Vanilla JS for the frontend; caching and streaming layers for snappy response + audio playback.
In short: local knowledge + retrieval (the brain), GPT-5 reasoning (the voice in the mind), and neural TTS (the voice you hear) all combine to make JI feel responsive and trustworthy.
- Context-first, no hallucinations: if it’s not in the profile, JI says it doesn’t know.
- Fast, lightweight backend (FastAPI) with a tiny RAG (Retrieval-Augmented Generation) pipeline.
- Voice output with caching for snappy playback.
- Programmed by Jason (the real me).