GenAI Portfolio Kit - 5 Production-Grade Projects (Source + Build Guides)

Stop building to-do-list clones. Ship the 5 projects that actually get GenAI interviews.

Five complete, working applications - RAG search, an autonomous agent, a real-time voice AI, a browser-testing agent, and a full-stack AI SaaS - each with real source code, a step-by-step build guide, and real evaluation numbers. Built entirely on free-tier APIs (Groq, Google Gemini free tier, local models), so you pay ₹0 to build and run every one of them.

What you get

• 5 complete codebases, zipped and fully self-contained - unzip, install, run
• 5 build guides(GUIDE.md) - not tutorials, teaching docs written from the actual code, explainingwhyevery architectural decision was made
• 5 READMEs with exact run commands,.env.examplefiles, and one-command local setup
• Real eval numbers for every project - no invented benchmarks, every metric is reproducible
• A portfolio overview guide mapping all 5 projects' stacks and the skills each one proves
• Zero-cost to build: every model call routes through Groq's and Gemini's free tiers, plus local CPU models - no credit card needed anywhere

The 5 projects

1. DocuMind - RAG Document Intelligence Ask questions over your own PDFs and get answers with clickable page citations. Ships the retrieval pattern most tutorials skip: dense + BM25 hybrid search -> RRF fusion -> cross-encoder reranking. Hit@5 96% -> 100%, MRR +0.138 measured before/after hybrid+rerank.React · FastAPI · Qdrant · Gemini

2. TaskPilot - Autonomous Research & Action Agent Give it a goal; it plans, searches, reads, critiques its own draft, and acts - pausing for your approval before anything risky. Includes a custom-built MCP server and durable checkpointing that survives a server restart mid-task.LangGraph · MCP · Groq · React

3. VoiceCoach - Real-Time Voice AI Interview Coach Talk to it. It interviews you out loud, scores your answers against a rubric, and coaches you back - with true barge-in (interrupt it mid-sentence like a real conversation). p50 end-to-end voice latency: 3.6 seconds. STT word-error-rate: 5.7%.WebSocket · Whisper · Kokoro TTS · Groq

4. AutoQA - Autonomous Browser QA Agent Describe a test in plain English; it drives a real Chromium browser, checks the flow, and files a structured bug report with screenshot evidence - with safety rails (allowlists, human approval before form submits) built in from day one.Playwright · Gemini Vision · LangGraph

5. SpendLens - Multimodal Expense SaaS (Capstone) Upload a receipt photo, get validated structured data with confidence scores, correct it in a review queue, then chat with your spending - grounded, cited, multi-tenant. Includes a CI pipeline that blocks bad pull requests on quality regressions. Next.js 16 · Supabase · Clerk · Vercel AI SDK

Who this is for

• Developers who want 5 resume-ready GenAI projects without spending weeks figuring out architecture from scratch
• Anyone prepping for AI engineer / agent engineer interviews who wants concrete talking points backed by real numbers, not buzzwords
• Students who want tolearn by reading real code, not another "call the OpenAI API in 10 lines" tutorial

Why this is different

Most GenAI tutorials stop at "call the model, print the response." Every project here handles what actually separates a demo from a portfolio piece: hybrid retrieval instead of naive top-k, human-approval gates before an agent writes anything, safety rails before letting an LLM drive a browser, confidence-gated human review instead of trusting a vision model blindly, and an eval suite with numbers you can reproduce yourself.

What you need

Free accounts only - Groq, Google AI Studio (Gemini), and (for project 5) Clerk + Supabase. No paid API, no GPU, no credit card.

Price: ₹199

One-time purchase. Instant download - 5 zips + all guides