Highlights
What I built
- URL discovery pipeline for department listing pages.
- Local LLM drafting option to keep sensitive content off cloud services.
- Review artifacts generated before any send action.
← Back to Projects
Machine LearningIterating
LabReach AI
Research outreach copilot: scrape, summarize, draft, review, then send.
LabReach started as a script to reduce repetitive outreach prep. I kept a human-review checkpoint as a hard requirement so automation helps without blindly sending.
- Date
- 2026
- Signal
- ML + Tools
- Build stage
- CLI pipeline working, expanding campaign tooling
- Stack
- Python, Ollama
ml-toolsautomationscrapingworkflow
Workflow preview
Project notes
Architecture
How the system works
- Scrape/extract modules feed structured profile context.
- LLM stage drafts outreach text from context templates.
- Campaign layer stores history and controls send scheduling.
Challenges
What made it hard
- JS-heavy faculty pages broke simple scraping logic.
- Email extraction quality varied across lab websites.
- Automation required strict safeguards around send behavior.
Lessons
What I learned
- Speed without brakes is not useful in outreach workflows.
- A review layer is both an ethics and quality requirement.
Stack / materials
PythonOllamaBeautifulSoupPlaywrightSQLiteGmail API
- Discovery + review tooling now matters more than single-email generation.
- Next step is stronger campaign observability and retry handling.
Media timeline
Build photos, clips, and process visuals. The goal is to show how the project evolved, not just the final screenshot.
Build snapshot
Iteration snapshot
Gallery 3