Skip to content
TARUN S / PORTFOLIOTAP OR PRESS ANY KEY TO SKIP

Developer

000

Dynamic FAQ Builder

TYPE
Full-stack search platform
STACK
12 COMPONENTS
STATUS
SOURCE AVAILABLE
FEATURED
—
LINKS
SOURCE ↗

01 / OVERVIEW

An admin-managed FAQ system where a custom TF-IDF relevance engine and Gemini AI work together to find and generate the best answer to any question.

Built as a team project (Team Tesseract) for PES University's Software Engineering course, Dynamic FAQ Builder is a full-stack platform that lets admins manage question-answer pairs while users search them through a custom-built relevance engine rather than simple keyword matching. I worked across the stack: a React + MUI frontend, an Express/Node REST API, Firebase Firestore for persistence, and a hand-written TF-IDF-style search service that weights matches by field (keywords 5x, question 3x, answer/tags 1x) and blends in a popularity signal from views and helpful votes. On top of that sits a Gemini-powered layer that can answer questions with no matching FAQ, generate answer drafts and tags for admins, and suggest related questions — with the API layer hardened with JWT auth, brute-force lockouts, rate limiting, Helmet CSP, and input sanitization against XSS.

GitHub ↗

02/ STACK & ARCHITECTURE

FRONTEND

  • React
  • React Router
  • Material UI
  • Vite
  • Tailwind CSS

BACKEND

  • Node.js
  • Express
  • Firebase Firestore
  • Firebase Auth
  • Google Gemini API
  • JWT
  • bcryptjs

03 / WHAT WAS ACTUALLY HARD

  1. 01

    Custom weighted relevance search engine

    Built a TF-IDF-based ranking service from scratch (no external search library) that tokenizes and scores FAQs per field — a 5x boost for keyword matches, 3x for question matches, 1x baseline for answer and tag matches — then blends in a popularity boost from view counts and helpful/not-helpful votes. Also supports fuzzy matching via Levenshtein distance and live search-as-you-type suggestions.

  2. 02

    Gemini AI answer generation and fallback

    When a user's question doesn't closely match an existing FAQ, the AI search endpoint hands the top-ranked candidates to Gemini, which synthesizes a direct answer from that context or, if nothing relevant exists, generates a general response. Responses are cached in memory to cut redundant API calls.

  3. 03

    Admin authoring tools powered by AI

    The admin dashboard lets staff create, edit, and delete FAQs, with Gemini assisting by drafting answer suggestions, improving existing answers, auto-generating tags, and categorizing questions — plus a dashboard view of total FAQs, category distribution, and top-viewed entries.

  4. 04

    Hardened Express API

    JWT-based admin auth with brute-force lockout after repeated failed logins, per-route rate limiting (tighter on search, looser on admin CRUD), Helmet CSP headers, express-validator field rules, and a custom sanitizer that strips script/iframe/img tags and inline event handlers from submitted content.

  5. 05

    Real-time Firestore sync and role-based access

    FAQs and AI Q&A history are stored in Firestore with live onSnapshot subscriptions so admin and user views stay in sync without polling. Firebase Auth handles user signup/login, with a role field on each user document gating access to the admin dashboard.