AI-Powered Customer Support Assistant
A customer support tool where every incoming query is answered using retrieval-augmented generation against a knowledge base, then automatically categorized and prioritized — with an admin dashboard for monitoring every conversation.
Problem
Support teams re-answer the same categories of question constantly, and without structure (category, priority, confidence) every incoming query requires a human to triage it before anyone even starts solving it.
Solution
Built a RAG-based assistant that answers from a knowledge base and simultaneously classifies and prioritizes the query, so the support workflow gets both an answer and the triage metadata needed to route it — visible to admins through a dedicated dashboard.
Architecture
- 01Flask backend with JWT auth and bcrypt password hashing for registration/login
- 02Knowledge base text is extracted, chunked, embedded with Jina AI, and indexed in FAISS
- 03Incoming queries are embedded and matched against the FAISS index for relevant context
- 04Retrieved context plus the query is sent to Groq's Llama 3.3 70B, returning answer, category, priority, and confidence
- 05PostgreSQL (Neon) stores users, conversations, and chat history
- 06Separate customer-facing and admin-facing frontends (HTML/CSS/JS)
Implementation
- ▸Built the JWT auth flow with secure password hashing before wiring up any AI functionality
- ▸Implemented the embedding → FAISS retrieval pipeline against the support knowledge base
- ▸Designed the Groq prompt to return structured output (category, priority, confidence) alongside the answer in one call
- ▸Built the admin dashboard as a separate view over the same conversation data for oversight
Challenges
- ▸Getting the LLM to reliably return structured metadata (category, priority, confidence) alongside a natural-language answer in a single response
- ▸Keeping retrieval relevant enough that priority/category predictions stay consistent across similar queries
Result
A live, deployed support assistant that answers, classifies, and prioritizes customer queries in one pipeline, with an admin dashboard for oversight.
More projects
MCPify
Analyzes any AI agent's URL, detects its framework, and generates ready-to-use MCP configs — and exposes itself as an MCP server.
NirmanAI 2.0
An AI agent for Indian home construction planning — RAG-grounded chat plus deterministic engineering calculators for BOQ, Vastu, and loan schedules.
IntelliDocs
A multi-user RAG platform: upload PDFs, ask questions, and get AI answers with page-level citations. Live and deployed.