SourceKit Talent Finder

AI-powered GitHub talent sourcing for technical recruiting. Find engineers by what they've actually built — their open-source contributions, repositories, and shipped code — not just their LinkedIn headlines.

What It Does

SourceKit turns GitHub's open-source graph into a recruiting pipeline:

• Research — Paste a job description or describe a role, and AI builds a sourcing strategy: target repos to mine, companies to source from, skills to weight, and EEA (Evidence of Exceptional Ability) signals to detect.
• Search — Finds real GitHub contributors matching your criteria, scores them 0–100, and surfaces hidden gems (high-quality, low-visibility engineers).
• Enrich — Looks up LinkedIn profiles via Exa semantic search and finds contact information.
• Pipeline — Kanban board to track candidates through Contacted → Not Interested → Recruiter Screen → Rejected → Moved to ATS.
• Outreach — AI writes personalized messages referencing the candidate's actual open-source work.
• Bulk Actions — Compare, rank, and draft outreach for multiple candidates with an AI chat interface.

Tech Stack

• Frontend: React 18, TypeScript, Vite, Tailwind CSS, shadcn/ui, Radix UI
• Backend: Supabase (PostgreSQL, Edge Functions, Auth)
• AI: Anthropic Claude (query parsing, candidate scoring, research, outreach generation)
• Search: GitHub REST API (contributor mining, profile enrichment), Exa (LinkedIn lookup)
• Payments: Stripe
• Hosting: Vercel
• Analytics: Vercel Web Analytics

Getting Started

Prerequisites

• Node.js 18+ (install via nvm)
• A Supabase project
• API keys for: GitHub, Anthropic, and Exa

Setup

# Clone the repo
git clone https://github.com/mrNLK/SourceProof.git
cd SourceProof

# Install dependencies
npm install

# Configure environment variables
cp .env.example .env
# Edit .env with your actual keys (see below)

# Start the dev server
npm run dev

Environment Variables

Copy .env.example to .env and fill in your values:

Variable	Purpose
VITE_SUPABASE_URL	Your Supabase project URL
VITE_SUPABASE_PUBLISHABLE_KEY	Supabase anon/public key
VITE_SUPABASE_PROJECT_ID	Supabase project ID
SUPABASE_SERVICE_ROLE_KEY	Supabase service role key (Edge Functions only)
GITHUB_TOKEN	GitHub personal access token
ANTHROPIC_API_KEY	Anthropic API key
EXA_API_KEY	Exa API key
PARALLEL_API_KEY	Parallel API key for Fast Entity Search and research
FINDEM_API_KEY	Findem signal import key
APOLLO_API_KEY	Apollo enrichment key
CLAY_API_KEY	Clay fallback enrichment key
MICROSOFT_*	Optional future Microsoft Entra/Graph OAuth and mailbox settings
SALESFORCE_*	Optional future Salesforce connected app and API settings
BD_*	BD operator identity, compliance footer, unsubscribe, and sending-domain settings
STRIPE_SECRET_KEY	Stripe secret key
STRIPE_WEBHOOK_SECRET	Stripe webhook secret
STRIPE_PRICE_ID	Stripe price ID

Important: The VITE_SUPABASE_URL and VITE_SUPABASE_PUBLISHABLE_KEY must also be set as environment variables in your Vercel project settings for the production deployment to work.

Deploying to Vercel

1. Connect this repo to a Vercel project.
2. Add all VITE_* environment variables in Vercel → Project Settings → Environment Variables.
3. Deploy. The vercel.json config handles SPA routing automatically.

SellKit BD Sourcing App

SellKit is a standalone business-development sourcing workflow at /sellkit. The legacy /bd-sourcing path also loads the same app. It uses the current Vite/React app, Supabase Postgres and Edge Functions, Vercel hosting, Exa Websets, Exa Company Search, Exa Agent, and Parallel server-side integrations. It does not introduce a separate backend or migrate the repo to Next.js.

V1 is manual-first because Microsoft Graph and Salesforce policy approval is unknown. The app can source, enrich, score, review, and prepare outreach without those integrations.

What is implemented:

• Deterministic scoring, human review gating, optional CRM/dedup helpers, target state machine, and template-bound email drafting.
• A login-gated review queue UI with Mariah onboarding saved to Supabase, signal evidence, verified-email status, review gate status, copyable email drafts, .eml download, CRM CSV export, and Sales Navigator note text.
• Supabase migration for bd_companies, bd_contacts, bd_signals, bd_targets, bd_outreach_touches, bd_email_messages, bd_salesforce_links, bd_suppressions, bd_scores, and bd_audit_logs.
• A bd-sourcing Edge Function with safe action routing and server-side Parallel Fast Entity Search at /v1beta/findall/entity-search.
• Server-side Exa Company Search for structured company seeds and Exa Agent for async multi-hop target discovery with structured output.
• Microsoft Graph and Salesforce write actions are optional future upgrades and remain blocked in v1.

Manual setup before first real run:

1. Rotate any API key that has been pasted into chat or logs.
2. Add PARALLEL_API_KEY and EXA_API_KEY to Supabase Edge Function secrets and Vercel environment variables. The browser never receives these keys.
3. Add Apollo and Clay keys, then confirm the exact people-match and email-verification fields.
4. Add Findem credentials or export/import flow for hiring, executive-change, and funding signals.
5. Fill BD_PHYSICAL_ADDRESS, BD_UNSUBSCRIBE_BASE_URL, BD_OPERATOR_NAME, and BD_OPERATOR_EMAIL.
6. Sign in to /sellkit, then complete the onboarding prompts for target companies, buyer titles, offer, buying signals, and email voice.
7. Keep BD_INTEGRATION_MODE=manual, APOLLO_SEQUENCE_ENABLED=false, and INSTANTLY_SEQUENCE_ENABLED=false until integrations are explicitly approved.

No CRM/customer/prospect suppression upload is required for v1 because a human approves every draft before any handoff or export.

Optional future setup, only if company policy allows it:

• Register a Microsoft Entra app for Microsoft Graph OAuth to create Outlook drafts or send through M365.
• Create a Salesforce connected app to run live dedup and write Leads/Contacts/Tasks.

Compliance defaults:

• No automatic email send, Outlook draft creation, Salesforce write, or Sales Navigator action executes in v1.
• Approved email is copied/downloaded for manual sending. Approved CRM records export as CSV for manual import.
• LinkedIn and Sales Navigator are manual only. The app prepares note text and profile links; it does not log in, scrape, or automate LinkedIn.
• Emails require verified work email, verified signal evidence, physical address, and one-click unsubscribe.

Scripts

Command	Description
npm run dev	Start dev server (localhost:5173)
npm run build	Production build
npm run preview	Preview production build locally
npm run lint	Run ESLint
npm test	Run tests (Vitest)

Project Structure

src/
├── components/      # UI components (search, pipeline, research, bulk actions)
├── hooks/           # Custom React hooks (subscriptions, etc.)
├── integrations/    # Supabase client and type definitions
├── lib/             # API helpers, utilities, and scoring (eea.ts)
├── pages/           # Route-level page components
├── services/        # External service integrations (Exa/Websets)
├── types/           # TypeScript type definitions
└── data/            # Static data and constants
supabase/
└── functions/       # Supabase Edge Functions (search, enrich, research, etc.)

Documentation

• Tester Guide — Full walkthrough of every feature, search tips, and workflow diagrams.
• Security Policy — How secrets and API keys are handled.

License

This project is source-available. The code is publicly visible for transparency and review, but no license is granted for redistribution or commercial use. All rights reserved by the author.