Aarav Jain
All work

CWW Group, AI Engineering Intern, Jul 2026 – now

Acquisition Agent

3 minutes, not 3 hours.

1 min read

My role
Built it: pipeline design, prompts, research layer, tests, MVP stabilization
Where
CWW Group, AI Engineering Intern
When
Jul 2026 – now
Stack
  • LangGraph
  • AWS Bedrock
  • Claude Sonnet 4.5
  • Claude Haiku 4.5
  • Tavily
  • LangSmith
  • pytest
GitHub

Run it.

A replay of real runs from the repo’s sample reports. Click any node to see what it does.

Acquisition research agent

Input

target="AC repair services in Los Angeles, CA"

in parallel

Weighted score

Not yet run

One line in, a diligence memo out

Eight LangGraph nodes scope the deal, research the market, run four analysts at once, then write and grade a private-equity memo. Press Run, or select any node to see what it does.

Replay of a real run. Order and outputs are real; time is compressed from about 3 minutes, and the analysts' finishing order is illustrative.

This run's report

  1. The problem

    Preflop

    Sizing up a small business for acquisition took an analyst 2–3 hours of searching, reading and spreadsheet work.

  2. The approach

    Flop

    A LangGraph pipeline: scope the thesis, run 10 M&A-focused searches, extract 16 structured fields, then fan out to four analysts in parallel (viability, build-out, profitability, competition) before synthesis and a self-evaluation pass.

  3. The hard part

    Turn

    Making it trustworthy: domain-specific queries instead of generic ones, isolated checkpoints per run, a simulated-data fallback, 26 unit tests, and every run traced and evaluated in LangSmith.

  4. What shipped

    River

    A PE-style diligence memo: investment thesis, Porter’s Five Forces, normalized EBITDA, deal structure and a weighted scorecard.

The result, or as poker players say, the showdown

~3 minutes vs 2–3 hours by hand. MVP stabilized and shipped. Sample run: AC repair in Los Angeles → 6.2 / 10, proceed with caution.