Autonomous engineering intelligence and career targeting system that discovers, scores, and investigates high-leverage open-source issues across high-growth AI startups.
The Startup Opportunity Radar is an end-to-end engineering pipeline built to solve a specific career problem:
How does a talented solo engineer systematically identify real, high-value technical problems in promising startup codebases, solve them, and use those contributions as proof-of-work to earn a high-paying USD role or paid collaboration?
Instead of guessing or browsing GitHub manually, the Radar builds a multi-phase intelligence pipeline:
YC Open Source Universe (Phase 4)
β
βΌ
Company β GitHub Entity Resolution (Phase 1)
β
βΌ
Public Repository Enumeration & Quality Scoring (Phase 0)
β
βΌ
Deterministic Issue Mining & Opportunity Scoring (Phase 2 & 2.1)
β
βΌ
Strategic Career Target Ranking (Phase 5)
β
βΌ
Code-Grounded AI Engineering Investigation (Phase 3)
β
βΌ
Actionable Build Brief & Targeted Upstream Pull Request
Startup Opportunity Radar
β
βββ Phase 0: Repository Discovery & Quality Scoring
β βββ Multi-query GitHub search orchestrator
β βββ Deterministic noise/tutorial/boilerplate filter
β βββ 4-dimensional scoring (Activity, Quality, Product Likeness, Opportunity Potential)
β
βββ Phase 1: Company Entity Resolution
β βββ Links startup brands (e.g. "Steel", "E2B") to GitHub orgs (steel-dev, e2b-dev)
β βββ Multi-signal domain matching, homepage normalization, and login heuristics
β βββ Strict confidence tagging (HIGH, MEDIUM, LOW, REJECTED, NOT_FOUND)
β
βββ Phase 2 & 2.1: Deterministic Engineering Opportunity Mining
β βββ Issue taxonomy classification (MCP, AI SDK, Agent Framework, Performance, Security)
β βββ Maintainer participation signals and PR collision context
β βββ Feasibility estimation (days, solo-engineer boundary clarity)
β βββ Multi-dimensional opportunity scoring (Proof-of-Work, Issue Quality, Feasibility, etc.)
β
βββ Phase 3: Code-Grounded AI Engineering Investigator
β βββ AST & file-level codebase retrieval via GitHub API / local tree
β βββ Strict fail-closed grounding guardrails (zero hallucinated filepaths/symbols)
β βββ Powered by Google Gemini Interactions API with bounded retries and sanitized logging
β βββ Actionable Build Brief synthesis for high-confidence BUILD recommendations
β
βββ Phase 4: Curated YC Startup Ingestion
β βββ Respectful, cached ingestion of YC Open Source startup directory
β βββ Complete company target normalization (batch, headcount, domain, subindustry)
β βββ Deduplicated SQLite persistence
β
βββ Phase 5: Career Target Ranking
βββ Evaluates expected return on engineering effort for proof-of-work strategy
βββ Combines Technical Opportunity Density (25%), Open-Source Quality (20%),
β Contribution Accessibility (15%), and Founder/Team Accessibility (15%)
βββ Target classification tiers: S (Immediate Priority), A (Strong), B, C, D
- Fact vs. Inference vs. Explicit Unknown:
- Facts: Direct data from GitHub API / YC Directory (stars, open issues, exact commit timestamps, license SPDX, team size).
- Inferences: Deterministic algorithms derived from facts (opportunity score, career target score, activity tier).
- Unknowns: Attributes unverifiable from public APIs (venture round size, code review latency) are explicitly marked as
UNKNOWNrather than guessed.
- Fail-Closed Code Grounding:
- The AI Investigator never claims an implementation root cause if the target code file was not retrieved. Unretrieved hypotheses are labeled
[HYPOTHESIS].
- The AI Investigator never claims an implementation root cause if the target code file was not retrieved. Unretrieved hypotheses are labeled
- Non-Destructive Persistence:
- Successful investigation reports and verified company mappings in SQLite are preserved and never overwritten by temporary network or validation failures.
- Python 3.11+
- Git
- (Optional) GitHub Personal Access Token (increases GitHub API limit from 60 to 5,000 req/hr)
- (Optional) Google Gemini API Key (for Phase 3 AI Codebase Investigation)
# 1. Clone the repository
git clone https://github.com/Priyavart/startup-opportunity-radar.git
cd startup-opportunity-radar
# 2. Create and activate a virtual environment
python -m venv .venv
# On Windows (PowerShell):
.venv\Scripts\Activate.ps1
# On macOS/Linux:
source .venv/bin/activate
# 3. Install dependencies in editable mode
pip install -e .
# 4. Configure environment variables
cp .env.example .env
# Edit .env and add your GITHUB_TOKEN and GEMINI_API_KEYThe radar provides a unified CLI via radar:
# View Top 20 Career Targets ranked for proof-of-work leverage
radar career-targets --top 20
# Inspect a specific company's career target profile
radar career-target "Browser Use"
# Export ranked career targets to JSON or CSV
radar career-targets export --format json
radar career-targets export --format csv# Ingest all YC Open Source companies with GitHub entity resolution
radar source yc --resolve-github
# List all stored YC companies
radar source yc companies --limit 50
# Export ingested YC dataset
radar source yc export --format json# Resolve a company name to its verified GitHub organization
radar company "Browser Use"
# Resolve multiple companies from a text file
radar company --file companies.txt# Mine and score all open issues across a company's repositories
radar mine-company "Browser Use"
# View top deterministic opportunities across all indexed repositories
radar opportunities --top 10 --tier S
# Inspect a single opportunity in detail with why-worth vs why-not-worth building
radar opportunity browser-use/browser-use#5252# Run deep technical codebase investigation on an issue
radar investigate browser-use/browser-use#5252
# Force refresh code retrieval and re-run AI evaluation
radar investigate browser-use/browser-use#5252 --refresh-ai# Run multi-query GitHub repository discovery
radar discover --queries 5 --min-stars 50
# View highest scoring discovered repositories
radar top --limit 20
# Inspect repository metadata, licensing, and activity signals
radar inspect browser-use/browser-useGit-o-hub/
βββ config.yaml # System queries, scoring weights, and cache TTL
βββ pyproject.toml # Package specification and dependencies
βββ .env.example # Template for API keys
βββ .gitignore # Production-ready git ignore
βββ data/
β βββ .gitkeep # SQLite database storage directory (data/radar.db)
βββ exports/
β βββ .gitkeep # Directory for JSON/CSV report exports
βββ src/
β βββ radar/
β βββ cli.py # Rich + Typer CLI interface
β βββ config.py # Pydantic Settings & YAML loader
β βββ db.py # SQLite schema manager & WAL connection pool
β βββ models.py # Unified Pydantic models for all 5 phases
β βββ ai/ # Phase 3: Code retriever, Gemini provider, investigator
β βββ career/ # Phase 5: Attractiveness scoring, eligibility, service
β βββ discovery/ # Phase 0: Search orchestrator, filter, classifier
β βββ entity/ # Phase 1: Candidate generator, evidence matcher, resolver
β βββ github/ # HTTPX client with rate limiting, cache, and retries
β βββ mining/ # Phase 2: Issue mining, feasibility, maintainer signals
β βββ scoring/ # Phase 0 & 2.1: Multi-dimensional scoring engines
β βββ storage/ # SQLite persistence layers for all entities
β βββ yc/ # Phase 4: YC directory client, parser, normalizer
βββ tests/ # 93 automated pytest unit & integration tests
The test suite contains 93 unit and integration tests covering all phases, scoring models, rate limiting, and failure handling:
# Run full test suite
pytest -v
# Run specific phase tests
pytest tests/test_career_ranking.py -v
pytest tests/test_ai_investigator.py -v
pytest tests/test_mining.py -v
pytest tests/test_entity_resolution.py -v
pytest tests/test_yc_source.py -vBased on the Radar's deterministic ranking and practicality filter, the top repositories to clone and contribute to are:
| Priority | Repository | Tech Stack | Domain | Why Clone |
|---|---|---|---|---|
| #1 | browser-use/browser-use |
Python 3.11+, Playwright | AI Agent Infrastructure | Massive traction (111kβ ), 7-person team with direct founder review, clean local pytest setup. |
| #2 | better-auth/better-auth |
TypeScript, Node.js | Developer Tooling & Auth | Exploding ecosystem (30kβ ), 4-person team, zero native binary dependencies, instant Vitest execution. |
| #3 | activepieces/activepieces |
TypeScript, Node.js | MCP & Workflow Automation | Leading MCP server repository (~400 tools), modular piece SDK, active maintainers. |
| #4 | cactus-compute/needle |
Python, PyTorch | Edge AI & Model Optimization | 14MB edge foundation model, cutting-edge token optimization, 8-person startup team. |
| #5 | beam-cloud/beta9 |
Go 1.22+, Docker | Serverless GPU Infrastructure | Distributed systems runtime, container sandboxes, high-value cloud engineering proof-of-work. |
This project is licensed under the MIT License β see the LICENSE file for details.