Lexicon Partners

A high-profile antitrust matter required reviewing 4.2 million documents spanning emails, contracts, presentations, scanned exhibits, and video depositions. The firm's existing eDiscovery platform could only search text, forcing the review team to manually examine 1.8 million image-heavy and multimedia documents that keyword search could not penetrate. At a projected review cost of $8.4M and 14-month timeline, the client was pushing for a more efficient approach.

Mixpeek indexes the full document universe across all modalities: OCR for scanned documents, visual analysis for charts and diagrams, speech-to-text for video depositions, and semantic embedding for email threads. Attorneys use a retriever-powered review interface to search by concept rather than keyword, finding relevant documents across all formats with a single query. Taxonomy enrichment auto-codes documents for privilege, relevance, and issue categories.

The 4.2 million document corpus was ingested through a parallel batch pipeline over 10 days. The review team started with a seed set of 500 known-relevant documents to calibrate retriever relevance scoring. Iterative feedback improved recall from 78% to 93% within the first two weeks. The solution integrated with the firm's existing Relativity instance through a custom connector that syncs coding decisions bidirectionally.