Skip to content

AI Playground

Ask this site about Rithesh.

A miniature retrieval system built the same way I build production RAG — an index, a scorer and cited sources. It answers from the résumé only, so it cannot invent a job he never had.

ask-rithesh · local retrievalonline

I'm a retrieval engine running over Rithesh's own résumé data — keyword scoring across projects, skills, roles and certifications. No external model, no API key, no cost per message. Ask me anything about his work.

Under the hood

How this thing actually works

Nothing is hidden — the retriever is a plain TypeScript module you could read in five minutes.

Indexed at build time

Every project, skill group, role and certification is flattened into 26 keyword-tagged documents when the app builds.

Scored, not guessed

Your question is tokenised, stop-words dropped, then each document is scored on keyword hits and prefix matches. Top three win.

Zero cost, zero latency

No model call, no API key, no vendor. The same shape as a RAG retriever — just with the generation step left out on purpose.

// lib/knowledge.ts — the whole retrieval step

const scored = docs
  .map((doc) => ({ doc, score: score(doc, tokenize(question)) }))
  .filter((s) => s.score > 1.5)
  .sort((a, b) => b.score - a.score)

return { answer: scored.slice(0, 3), sources: [...] }

// swap in an LLM by passing these documents as grounding context

What's next

Got something that needs building?

Open to full-time roles & select freelance work. Send a brief, a rough idea, or just say hello — I answer every message.