# TwinlyAI

AI SaaS, 2025–26.

Resumes become AI twins recruiters can interview, backed by RAG search.

## Objective

A resume reduces a person to keywords, and recruiters have too many to read. We built a way for recruiters to question a candidate's AI twin before booking a call, and for candidates to show how they think, not only what they did.

## Approach

- The resume is parsed, split and indexed in a vector database.
- The twin answers only from what it retrieves, and streams its reply.
- Candidates can connect GitHub so the twin can talk about their real code.
- Recruiters search candidates by meaning rather than keywords, and developers get API keys to connect it to their hiring tools.

## How the system fits together

Parts:
- Candidate (build layer, person)
- Resume upload (build layer, surface)
- GitHub sync (build layer, channel)
- Parse & embed (intelligence layer, engine)
- Vector database (intelligence layer, store)
- AI twin chat (intelligence layer, engine)
- Recruiter (build layer, person)
- Semantic search (intelligence layer, engine)
- API keys (build layer, channel)

Connections:
- Candidate → Resume upload
- Candidate → GitHub sync
- Resume upload → Parse & embed
- GitHub sync → Parse & embed
- Parse & embed → Vector database
- Vector database → AI twin chat
- Recruiter → AI twin chat
- Recruiter → Semantic search
- Semantic search → Vector database
- API keys → AI twin chat

## Features

### A twin grounded in the resume

Recruiters ask questions; answers come from the candidate's own material and stream in as they're written.

### GitHub sync

Repositories are synced into the twin's knowledge, so it can discuss real projects.

### Search by meaning

Recruiters find candidates by what they've done, not by matching keywords.

### API access

Teams can connect twins to their own hiring tools.

## Craft

- Retrieval augmented generation over a vector database, with streamed responses.
- Three sign in providers and a GitHub connector.
- Walkthrough video sized to the screen; reduced motion respected.

## Results

In private development. We'll share usage once it's measured.

## Credits

Credits: the team. Disciplines: Applied AI, Full stack engineering, Product design.
Stack: Next.js, React, Tailwind CSS, FastAPI, MongoDB, Qdrant vector database, AWS.

Source: https://magrus.vercel.app/work/twinlyai
