A CrewAI-based research pipeline that collects company intelligence and creates a personalized outreach sequence through a four-agent workflow.
The project is being upgraded from a notebook prototype into an evidence-grounded application. See UPGRADE.md for the roadmap.
CLI input
-> lead discovery
-> email generation
-> workflow review
-> quality validation
-> timestamped Markdown result
Application code lives in src/ai_research_agent. The original crew1.ipynb remains available as a reference and demonstration, but it no longer starts a live API run automatically.
- Python 3.10 or newer
- A Groq API key and compatible Groq model name
- Optional Serper API key for search
python -m venv .venv
.venv\Scripts\activate
python -m pip install -e .
copy .env.example .envEdit .env and provide GROQ_API_KEY and GROQ_MODEL. Never commit .env.
Validate configuration without making external calls:
research-agent doctorRun the workflow:
research-agent run \
--company "DeepLearning.AI" \
--industry "Online education" \
--decision-maker "Andrew Ng" \
--position "Founder" \
--milestone "a recent AI education initiative"Use research-agent --help to see all options. Results are written to outputs/ unless --output specifies another path.
| Variable | Required | Purpose |
|---|---|---|
GROQ_API_KEY |
Yes | Authenticates model requests |
GROQ_MODEL |
Yes | Selects the model explicitly |
SERPER_API_KEY |
No | Enables Serper-backed search |
RESEARCH_OUTPUT_DIR |
No | Changes the output directory |
RESEARCH_MAX_RPM |
No | Controls request rate; defaults to 5 |
RESEARCH_VERBOSE |
No | Enables verbose CrewAI output |
python -m compileall src
python -m ai_research_agent --helpThe next milestone adds real lead scoring, structured evidence schemas, and comprehensive tests.
Generated research can be incomplete or incorrect. Verify decision-maker information, claims, and outreach content before use. Respect privacy, applicable anti-spam law, and website terms.
MIT. See LICENSE.