Fund-of-Funds LP-Reporting Automation
Automated the quarterly LP-reporting workflow for a $14B AUM fund-of-funds. Client named on request.
Sat Mar 15 2025
A $14B AUM fund-of-funds, named on request.
The problem
Quarterly LP reporting at a fund-of-funds is a deceptively heavy workflow: capital account statements, underlying-manager reports, and performance data arrive in inconsistent formats and scattered across systems. The hard part isn't formatting — it's verifying the actual numbers across all those sources, reconciling them, and rolling them up correctly for each limited partner. It's high-stakes, deadline-bound, and almost entirely manual.
What I built
An agentic LangGraph pipeline that ingests the inbound fund documents, verifies the actual figures across the scattered sources, reconciles them, and writes the verified results straight back into the client's SharePoint — no human in the loop.
- Number verification, not re-templating — it validates the real values against every source they appear in, rather than just reshaping a template.
- Orchestrated runs + mixture of models — a LangGraph flow that fans work out across models and stages.
- Adversarial review for triangulation — models check each other's extractions so a figure has to survive cross-examination before it's trusted.
- Write-back into SharePoint so the updates land exactly where the team already works.
How I got there
I tested roughly 14 different architectures. Sample precision was perfect; recall landed around 80–90%. The trade-offs were stark — some configurations ran as 8–9-hour overnight jobs, others finished in ~10 minutes. I settled in the middle, balancing accuracy, quality, speed, and cost: a LangGraph flow with orchestrated runs, a mixture of models, and adversarial reviews that triangulate every figure.
Outcome
- Compressed weeks of manual work into a single day.
- Zero humans in the loop — extraction, verification, and SharePoint write-back run end to end.
- Quarterly LP reporting becomes a reviewed, repeatable run instead of an all-hands scramble.
Why it matters
LP reporting is exactly the kind of high-value, low-glory workflow agentic AI should own: bounded, repetitive, expensive in senior time — and unforgiving about accuracy. Done right, it hands that time back to the investment team while making the numbers more trustworthy, not less.