AI · 2025 ·Creator · Prototype
AI Support Engine
A retrieval-augmented support assistant that ingests technical documentation and answers questions with cited, semantically-matched solutions — plus a feedback loop that learns which answers actually resolved the issue.
A
Overview
Coming from years of frontline technical support, I wanted to test a hypothesis: how much of escalation triage could an AI handle if it actually read the docs? The AI Support Engine is that experiment — a retrieval-augmented assistant that turns a pile of technical manuals into an answer engine.
What I built
- Document ingestion — chunks and embeds PDF knowledge bases into a persistent FAISS vector store for fast semantic search.
- Cited answers — retrieves the most relevant passages and composes an answer grounded in them, rather than hallucinating from thin air.
- A learning loop — tracks which suggested solutions users confirm as helpful and boosts them over time, while logging unresolved questions for a human to turn into new KB content.
Why it matters
It’s the natural marriage of my two worlds — support engineering and AI. The same instinct that makes a good runbook (capture what works, reuse it) becomes a system that gets smarter with every resolved case.
Built with Python Streamlit LangChain FAISS Sentence-Transformers
Next project
Support-Ansible →