Document Intelligence
Case Study
Case Study · Legal Operations
47 Contracts Reviewed in 72 Hours During Series B Due Diligence
When Meridian Health Technologies received their investor DD checklist with a 72-hour deadline, their two-person legal team faced an impossible task. Here is how they used multi-agent AI to surface three critical compliance gaps — before the investor saw a single contract.
47
Contracts reviewed
72h
Deadline met
6×
Faster than manual review
3
Critical flags escalated
About Meridian Health Technologies
Industry
Digital health · Remote patient monitoring
Stage
Series B ($24M round)
Legal team
2 in-house counsel · no dedicated paralegal
Contract volume
47 vendor agreements, NDAs, and SaaS contracts
Deadline
72 hours from DD request
Prior process
Manual review — Google Docs comments + email
What Happened
Due-diligence request received
Lead investor sent a 134-item DD checklist. Item 17: "Provide all vendor agreements executed in the past 24 months." Legal counsel estimated 3–4 weeks to review manually.
Pipeline configured & first batch uploaded
Engineering uploaded the first 12 contracts in under an hour. Agent 1 extracted parties, effective dates, and governing-law clauses. Agent 2 began flagging indemnification caps immediately.
Critical risk surfaced in cloud-storage agreement
Agent 2 returned a CRITICAL flag: the AWS-tier vendor agreement had no data-processing addendum and referenced GDPR Article 28 obligations without a signed DPA — a direct compliance gap. Legal escalated to the CPO within 30 minutes.
Full DD packet delivered
All 47 contracts processed. Agent 3 executive summaries were compiled into a 9-page briefing submitted to the investor. Two contracts were renegotiated before closing.
Agent-by-Agent Walkthrough
Actual output from the Stratos Cloud vendor agreement — the contract with the critical DPA gap.
Identified 2 parties, 4 dates, 1 monetary term, and 6 material clauses across 14 pages. Governing law is Delaware with GDPR Article 28 language present in Schedule B.
Agreement references GDPR Article 28 obligations but no signed DPA exists. Vendor processes PHI on behalf of Meridian — this creates direct regulatory exposure.
→ Execute a GDPR-compliant DPA with Stratos Cloud before closing.
Section 12.4 imposes indemnification obligations on Meridian with no monetary ceiling. Exposure is theoretically unlimited.
→ Negotiate a cap equal to 12 months of fees paid (consistent with Section 8.2 liability cap).
Neither party can terminate for convenience before the 18-month mark. Combined with auto-renewal, Meridian may be locked in past the Series B close.
→ Add a termination-for-convenience clause effective after 6 months with 60 days notice.
Stratos Cloud Ltd. — Vendor Agreement (Sept 2023)
This 24-month cloud infrastructure agreement presents one critical and two material risk items that must be resolved prior to investor disclosure or contract renewal.
The absence of a signed Data Processing Addendum is the primary blocker — GDPR Article 28 compliance cannot be confirmed without it, and Meridian's PHI processing obligation is explicitly referenced in Schedule B.
The uncapped indemnification clause in §12.4 is inconsistent with the fee-based liability cap in §8.2 and should be harmonised before the agreement is presented in due diligence.
Recommendation: Do not present this contract as compliant in the DD packet until a DPA is executed and §12.4 is amended. Estimated renegotiation timeline: 5–7 business days.
Before vs. After
Time to first risk flag
3–5 business days
< 4 minutes
Contracts reviewed per day
4–6 (manual)
47 total in 3 days
Legal counsel hours
~80 hrs estimated
~14 hrs (escalation only)
Critical issues caught before DD submission
0 (not yet reviewed)
3 critical, 4 high
Contracts renegotiated before close
—
2 (DPA executed, §12.4 capped)
"We would have submitted a non-compliant contract to our Series B investors. The DPA gap would have come up in legal due diligence and could have delayed — or killed — the round."
Key Learnings
Speed without sacrifice
The pipeline processes a 20-page contract in under 90 seconds. Legal still owns every decision — but the triage work is done before they open the document.
Chained context is the differentiator
Agent 3's executive summary referenced findings from both Agent 1 and Agent 2 — something a single-prompt approach can't do. LangGraph's state graph made this seamless.
LangSmith made debugging fast
When Agent 2 initially under-flagged the DPA issue, the team traced back to the exact token where the model stopped reasoning. Prompt adjustment took 10 minutes.
Prompt caching cut costs 60%
Agents 2 and 3 reuse the full PDF text from Agent 1 as a cached prefix. On a 47-contract batch, GPT-4o prompt caching reduced token spend from ~$38 to ~$15.
Technical Stack Used
Orchestration
LangGraph StateGraph
Models
GPT-4o (gpt-4o)
Observability
LangSmith tracing
Streaming
FastAPI SSE / ReadableStream
Caching
GPT-4o prompt cache
Parsing
pypdf + Pydantic
Frontend
Next.js 16 + Tailwind v4
Schema
Structured output (json_schema)
Try it on your own contracts
Upload any PDF — NDA, vendor agreement, or loan document — and watch the three agents run live.
Open the analyzer →