Orange

Your agentic hiring team.

Orange replaces the busywork of recruiting. Describe a role, drop in the CVs, and it reads every one, ranks the shortlist, and shows its work. Sourcing to offer, run by AI.

Early prototype. The core loop works end to end today. Everything under Roadmap is where we take it next.

What it does

You describe the role. Orange does the reading: it parses every résumé into a clean profile, scores each candidate against the job across skills, experience, education, and logistics, and writes out the reasoning behind every ranking. The scores are real and every one is auditable, with the model quoting the exact lines from the CV that justify it. No forms. No keyword filters. No black box.

How it works

Uploads land on a queue, so ranking runs in the background. The recruiter never waits on the model.

flowchart TD
    subgraph Client["Recruiter web app · Next.js"]
      direction LR
      A["Job chat"]
      B["Pipeline Kanban"]
      C["Candidate profile"]
    end

    Client -->|REST| API["API layer · route handlers (BFF)"]
    API -->|commands| ORCH["Orchestrator"]
    API -->|upload| OBJ[("Object storage · CV files")]
    API -->|reads| PG[("PostgreSQL + pgvector")]
    ORCH -->|enqueue| Q[["Job queue · async parse + rank"]]
    Q -->|consume| AI

    subgraph AI["AI agent service · FastAPI"]
      direction LR
      JA["Job Architect"]
      CV["CV Parser"]
      RK["Ranker"]
    end

    AI -->|LLM calls| LLM(["LLM API · Claude / OpenAI"])
    AI -->|read / write| PG

Loading

Roadmap

We start where hiring hurts most and expand until the whole pipeline runs itself.

• Read, score, and rank candidates with reasoning you can act on
• ATS sync: Greenhouse, Lever, Workday
• Inbound capture and outbound follow-ups over email and WhatsApp
• AI first-round interviews, scored and auto-advanced
• Orange in the room: a live copilot for human technical interviews
• Sourcing to offer, end to end

The goal: the hiring team, in software.

Stack

Lean now so it ships fast. Built to grow into a stack that scales.

	Now (prototype)	At scale
Web	Next.js · TypeScript · Tailwind	Next.js
API and services	Next.js route handlers (BFF)	Go services
Agents	Python · FastAPI	Python · FastAPI
Reasoning	Claude / OpenAI	Claude / OpenAI + tuned models
Vector search	Postgres + pgvector	dedicated vector DB (Qdrant / Pinecone)
Queue and cache	in-process queue	Redis + event stream
Data	PostgreSQL	PostgreSQL + read replicas
Infra	Vercel · Railway	AWS · Docker · Kubernetes

Run locally

Prerequisites: Node 22+, uv, and Docker.

git clone https://github.com/saadhtiwana/orange.git
cd orange
cp .env.example .env        # add your ANTHROPIC_API_KEY

1. Database — Postgres with pgvector, on port 5432:

docker compose up -d postgres

2. AI service — FastAPI on localhost:8000:

cd ai
cp ../.env.example .env     # needs ANTHROPIC_API_KEY
uv sync
uv run uvicorn app.main:app --reload

Check it: curl localhost:8000/health

3. Web app — Next.js on localhost:3000:

cd web
cp ../.env.example .env     # needs DATABASE_URL and AI_SERVICE_URL
npm install
npx prisma migrate deploy   # create the tables
npm run dev

Open localhost:3000 — it lands on the Job Architect. Describe a role and you get a structured job description back, saved to Postgres.

Working on it

# web/
npm run lint && npm run format:check && npm run typecheck && npm test && npm run build

# ai/
uv run ruff check . && uv run ruff format --check . && uv run mypy app scripts && uv run pytest

CI runs exactly these on every push and PR. Tests mock the LLM, so they need no API key and cost nothing.

After changing a contract in ai/app/contracts/models.py, regenerate the artifacts:

cd ai  && uv run python scripts/export_schemas.py
cd web && npm run gen:contracts

See docs/contracts.md for what the contracts are and CONTRIBUTING.md for the branch and commit conventions.

Team

@saahtiwana · @ahmadmustafa02 · @abdullahxdev

_{© 2026 Orange}