Message Navigation & Record Analysis
Chronology Reconstruction, Contradiction Detection & Source-Linked Record Navigation
Transforms fragmented communication archives into source-linked chronologies, surfaces factual contradictions, and enables rapid record navigation while strictly preserving provenance and context across high-stakes review environments.
What was difficult, fragmented, or broken?
In high-conflict legal proceedings, regulatory investigations, and compliance disputes, communication evidence is scattered across gigabytes of unindexed exports: thousands of iMessage/WhatsApp screenshots, carrier CSV dumps, audio transcripts, and third-party messages. Manual review takes hundreds of billable hours, while chronological fragmentation obscures coercive patterns, timeline gaps, and contradictory statements.
Why the obvious solution wasn't enough
Standard e-discovery tools are built for corporate email threads and struggle with mobile screenshots, burner numbers, non-sequential image dumps, and subtle contextual manipulation. Optical Character Recognition (OCR) tools often fail on mobile chat bubbles, while keyword searches miss critical behavioral patterns such as unanswered message bursts or conditional threats.
My Role
What I Built
- The `muse-forensics` (`mf`) CLI utility powered by `uv` for sub-second execution across massive local file trees.
- A native Windows UWP OCR pipeline that extracts high-accuracy text and timestamps from mobile screenshots without transmitting sensitive evidence to third-party cloud APIs.
- A PyMuPDF indexing engine capable of instantaneous multi-term regex querying across 5,000+ page exhibit binders.
- A message log aggregator compiling conversation velocity, response latency, nocturnal contact clustering, and multi-number attribution.
- A neutral event indexer transforming raw message transcripts into objective, chronological ledgers with direct page/citation coordinates.
$ mf ingest ./exhibit_binder_vol_3/ --ocr uwp --workers 8 [INFO] Discovered 4,218 screenshot exhibits and carrier logs [OCR] Native UWP OCR engine initialized (local-only pipeline) [INDEX] 10,482 message entities parsed across 14 timeline bands [AUDIT] 17 timeline discrepancies and 4 burner identities clustered [OUTPUT] Generated verified chronology ledger: exhibits_chronology.csv
How I Approached It
I treated communication analysis as a data provenance problem. Every extracted line maintains strict cryptographic linkability back to its original source page and timestamp. The system strictly avoids subjective emotional categorization, indexing events through neutral behavior types (e.g., conditional demands, resource withholding, timeline discrepancies) to ensure verifiable auditability.
AI is an instrument, not an authority.
Used AI reasoning models to develop adversarial regex test patterns for phone number obfuscation and date format parsing variations. Structured prompt pipelines to identify structural contradictions across testimony statements and contemporaneous text logs, requiring citations for every identified anomaly.
Key Decisions & Tradeoffs
Kept all data processing strictly local on-device (zero cloud transit) to uphold evidentiary chain of custody and attorney-client privilege.
Employed native Windows UWP OCR over Tesseract due to superior handling of variable DPI, modern mobile fonts, and chat bubble contrast levels.
The Result
Operational Python CLI toolkit (`mf`) that indexes thousands of pages in seconds, generating audit-ready CSV ledgers used in active legal and investigatory reviews.