Chopin: ICP Research

Chopin GTM creates a personalized sales brief for your ICP and triggers events along with multichannel outreach pack in seconds

Comment

Inspiration

GTM teams waste 2-3 hours per account doing repetitive research: scraping LinkedIn, reading press releases, crafting personalized messaging. We watched BDRs copy-paste the same templates, AEs scramble before calls, and SDRs burn out on low-response outreach. The problem isn't effort—it's that manual research doesn't scale. We built Chopin to turn company names into ready-to-use account intelligence in under 60 seconds.

What it does

Input a company name. Chopin outputs:

  • Account brief summarizing business model, recent activity, and GTM angles
  • 3 outreach variants (email, LinkedIn, call script) with personalized hooks
  • Call opener referencing real trigger events
  • Top 3 trigger events with dates (funding, product launches, exec hires)
  • One-page visual brief for internal handoffs

Everything is sourced from public data and tailored to the prospect's current situation.

How we built it

Orchestration stack:

  • Yutori generates research plans ("check funding history, find tech stack, scan news from last 90 days")
  • TinyFish executes the plan, scraping company websites, LinkedIn, Crunchbase, and news sources
  • Claude API (Sonnet 4) transforms raw data into personalized messaging and briefs
  • Freepik generates a formatted one-page visual summary
  • Retool hosts the UI with live run logs showing each agent's progress

Backend is FastAPI in Python. We used VS Code with Copilot for rapid prototyping, ChatGPT for debugging TinyFish API quirks, and Claude for refining agent prompts.

Challenges we ran into

Agent coordination: Getting Yutori's research plan to map cleanly to TinyFish's data sources required several prompt iterations. Initial attempts returned generic plans; we had to add example outputs to guide specificity.

Rate limits: TinyFish throttled us during testing. Solution: cached responses for common test companies and batched Claude API calls.

Retool constraints: Can't run Retool locally in an IDE. We built the backend API first, tested with curl, then connected the UI in the final 15 minutes.

Data quality: TinyFish sometimes returned stale data. We added a "freshness filter" to prioritize results from the last 6 months.

Accomplishments that we're proud of

Built end-to-end in 58 minutes. The trigger event detection works surprisingly well—it caught a Series B announcement we'd missed manually. The outreach messages pass the "would I respond to this?" test. Most importantly, the live run log makes the AI work transparent instead of a black box.

What we learned

Agent composition matters more than individual AI quality. Claude is excellent at writing, but without structured research from TinyFish, it hallucinates. Yutori's planning step seems redundant until you realize it prevents TinyFish from scraping irrelevant pages.

GTM teams want explainability. Early versions just showed final output. Adding the Retool log panel ("TinyFish found 3 funding events...") increased trust immediately.

Caching is critical for agentic systems. We spent 40% of dev time on re-running the same scrapes during debugging.

What's next for Chopin: ICP Research

CRM integration: Auto-sync account briefs to Salesforce/HubSpot, trigger regeneration when trigger events update.

Team playbooks: Let users define custom research templates ("always check AWS Marketplace reviews for infrastructure companies").

Multi-company mode: Batch-generate briefs for entire target account lists, export as CSV.

Feedback loop: Track which outreach messages get responses, fine-tune Claude prompts based on conversion data.

Expansion beyond sales: Adapt for investor due diligence, partnership research, competitive intelligence.

Built With

Share this project:

Updates