Built for Defensibility.
Every feature designed with lit-support professionals in mind. Precision search, automated privilege detection, and court-ready audit trails.
Search Intelligence
Multi-modal retrieval for maximum recall
Hybrid Search
DEFAULTBM25 lexical + semantic vectors combined via Reciprocal Rank Fusion. Catches both exact terms and conceptual matches.
RAG Fusion
ADVANCEDMulti-query expansion generates 4 search variants, runs parallel hybrid searches, then fuses results across all queries.
Full Pipeline
PRECISIONComplete pipeline with LLM reranking. Gemini 3 Flash scores and reasons about each candidate for maximum precision.
Accuracy validation in progress. We're running comprehensive benchmarks on the Enron corpus to provide verified precision metrics. Check back soon for validated accuracy claims.
AI Triage Engine
Gemini 3-powered document classification
Four-Way Classification
Every document receives a decisive classification with confidence scoring and mandatory source citations.
Citation Enforcement
offset: 1247-1298
Document Processing
Forensic-grade extraction pipeline
Container Extraction
ZIP, PST, MSG with full recursive extraction. Zip bomb protection via compression ratio analysis.
Document Processing
Native text extraction from PDFs, Office documents, and 100+ file types with metadata preservation.
Tiered OCR
Tesseract baseline with Gemini 'hard lane' for difficult pages. Automatic hardness scoring promotes bad OCR to LLM.
Email Threading
Header-aware chunking preserves conversation context. Attachments extracted and processed recursively.
Privilege Guard
Automated attorney-client detection
Catch privilege risks before they ever hit an external platform. Pattern-based detection combined with entity recognition flags potential attorney-client communications for human review.
Review Queue
Forensic Defensibility
Court-ready audit infrastructure
Content-Addressed Storage
SHA256 deduplication ensures identical files are stored once. Full provenance chain from source to artifact.
Immutable Audit Logs
Append-only, cryptographically chained logs. Every action timestamped and attributed for court-ready defensibility.
Citation Enforcement
Every AI decision MUST cite source text. No citations = forced human review. Zero tolerance for hallucinations.
Model Versioning
Every AI output tracked with model ID, version, prompt hash, and token counts. Full reproducibility.
Export to Your Stack
Generate load files compatible with every major review platform. Your data, your workflow.