Kensink Labs
LEGALTECH · DOCUMENT AIAffidavit Mapp (US)· US

Affidavit Mapp.
Court-ready bank statements, days to minutes.

99.7% data integrity.

99.7%
Data integrity vs. ground truth
OCR + LLM PIPELINELEGALTECH · DOCUMENT AI

Affidavit Mapp: Court-ready bank statements, days to minutes. 99.7% data integrity.

Family-law firms drown in bank statements. We built a secure OCR + LLM pipeline that turns raw PDFs into court-admissible reports in minutes, not days. It uses strict grounding, full chain-of-custody, and hits 99.7% data integrity measured against ground truth.

99.7%
Data integrity vs. ground truth
Days→min
Per-case processing time
100%
Outputs grounded to the source PDF
8 wk
Kickoff → production for first firm
01 · THE PROBLEM

Where they were stuck.

Family-law analysts were spending entire days per case reading bank statements line by line, building spreadsheets that would later be sworn into court. The variance between analysts was high; the audit trail was a folder of Excel files; and the bottleneck was the analyst, not the law. Existing 'AI document tools' produced confident-looking nonsense: hallucinated transactions, dropped pages, no verifiable source.

02 · OUR APPROACH

How we built it.

  • 01Layered OCR: open-source OCR engine for structure + a vision-LLM pass to recover degraded scans, falling back to manual review only for low-confidence pages
  • 02Retrieval-Augmented Generation grounded against the source PDF: every extracted transaction carries a page reference + bbox so the answer is verifiable back to the original
  • 03Strict eval suite of 1,200 manually-labeled statements as ground truth, run on every model change; production gated on field-level precision/recall thresholds
  • 04Chain-of-custody: SHA-256 hash of the source PDF, all model outputs signed with extraction timestamp, full audit log per case
  • 05Court-admissibility formatting: output post-processed into the exact report shape that judges accept, with no creative formatting
  • 06Confidentiality boundary: processing inside the firm's tenancy, source documents never leave their cloud
When opposing counsel can't poke a hole in your output, that's when you know the system works. We've been through depositions on this data with zero challenges sustained.
Research Director
Affidavit Mapp
[TECH STACK]
  • Python
  • OCR engine
  • Object storage
  • Multimodal LLM
  • Postgres
  • Eval framework
[ENGAGEMENT]
DurationMulti-engagement advisory (2024)
ClientAffidavit Mapp (US)
ShapeOCR + LLM PIPELINE
HandoffFull ownership · 90-day warranty
START YOUR OWN PROJECT

Bring a real problem.
We’ll bring code on day one.

← All casesCASE · AFFIDAVIT-MAPP