Perplexity Clone
2026Live
Cited AI answer engine with web-grounded answers, citations, threaded follow-ups, and an honest single-user V2 scope.

Overview
A Perplexity-inspired answer engine: ask a question and get web-grounded answers with citations and threaded follow-ups.
What I built
A full-stack monorepo:
- Backend: a NestJS modular monolith (TypeScript, PostgreSQL, Prisma) that owns the API, persistence, Tavily web-search integration, and provider-agnostic answer generation (OpenAI or Groq).
- Frontend: a Next.js + Tailwind responsive two-column thread UI with custom markdown parsing that renders citation badges.
- Docker Compose shapes a production-like local environment, and a separate deployment guide covers migrations, secrets, networking, and verification.
Challenges
V2 is deliberately single-user and local: no auth, no user-scoped data, no rate limits, and no multi-tenant guardrails; those are tracked as later productization work. That explicit scoping is itself an engineering decision.
Outcomes
Live demo reachable; listed on the approved master resume. No production/commercial claims are made.
Narrative sources:
docs/research/featured-project-research.md§2 + evidence ledger CL-13. Honest V2-scope framing kept.