AISupport Copilot

RAG-powered AI support

AI Support Copilot

A RAG-powered AI support chat system for any product — index your help documentation, answer with cited sources, stream replies live, and refuse when docs do not match.

For now, built for InterviewPro.info platform support.

Start chattingBrowse Help Center
Why trust itHow chat worksTry a questionTechnical guide
16
Help articles indexed
11
Gen AI topics built
Hybrid
Vector + keyword RAG
SSE
Live streaming chat

Live preview

mateshwari

Support · InterviewPro.info

Online
What is machine coding practice on InterviewPro.info?

mateshwari

Machine Coding includes 70+ JavaScript and 30+ React interview-style challenges — all in your browser with a live editor, test runs, and filters by company and difficulty. Your solved work syncs to your progress dashboard after sign-in.

Sources

Machine Coding practiceProgress dashboard

Why it works

Support you can trust

Grounded in real documentation — with the UX of a live help desk.

  • Retrieval-first

    Hybrid search finds Help Center articles before the model writes a single word.

  • Cited sources

    Every reply links to the articles used — verify details anytime.

  • Live conversation

    Streaming Markdown replies, follow-up memory, and mateshwari as your support specialist.

  • Safe by design

    Injection checks, confidence guardrails, and allowlisted tools for billing questions.

How chat works

Six steps from your question to a grounded, streamed reply.

  1. 1Client

    Send message

    SupportCopilot + Zustand POST question and chat history to /api/chat. API keys never leave the server.

    React · Zustand · lib/chatClient.js

  2. 2Server

    Safety check

    Server scans for prompt-injection patterns before any retrieval or model call.

    lib/security.js

  3. 3Server

    Hybrid RAG

    Vector embeddings + BM25 keyword search over Help Center chunks, merged and reranked.

    OpenAI embeddings · lib/hybridRetrieval.js

  4. 4Server

    Guardrail

    If top retrieval score is too low, the app refuses instead of guessing.

    lib/prompt.js

  5. 5Server

    Agent + tools

    ReAct planner decides: answer from docs or call get_subscription_status for billing.

    OpenAI tools · lib/tools.js

  6. 6Client

    Stream reply

    Grounded answer streams over SSE — Markdown in UI, source chips, token stats on hover.

    SSE · react-markdown · SupportCopilot.jsx

Try the live chat flow

What you can ask about

  • A2Z Frontend & Machine Coding
  • DSA by Patterns & Gen AI track
  • ResumePro & EvalPro
  • Progress dashboard & sign-in
  • Job Board & Interview Guide
  • Plans, billing & troubleshooting
Browse Help Center

Technology stack

Full-stack Gen AI support — modern frontend, server-side trust boundary, production-style RAG.

Frontend

  • Next.js 16

    App router, API routes, SSR

  • React 19

    Chat UI, streaming message list

  • Tailwind v4

    Landing, chat, docs styling

  • Zustand

    Chat state, history, stream events

  • react-markdown

    Rich assistant replies (ChatGPT-style)

AI & retrieval

  • OpenAI GPT

    Grounded answer generation + ReAct planning

  • text-embedding-3-small

    Semantic search over help articles

  • Hybrid RAG

    Vector + BM25 with reranking

  • In-memory index

    Chunked docs from supportDocs.js

Backend & trust

  • SSE streaming

    Token-by-token replies to the browser

  • ReAct agent

    Plan → optional tool → answer

  • Guardrails

    Injection block + low-confidence refusal

  • Golden evals

    npm run eval retrieval regression tests

Architecture diagrams and code paths in Technical guide

Gen AI curriculum

11 Gen AI topics — all implemented

Each topic is wired into this app with real code—not just listed on a resume. Open the technical guide for architecture notes and file paths.

Full guide →

Ready to try it?

Open the live copilot or explore the Help Center and technical guide.

Open chatHelp CenterTechnical guide