Equall Improves Legal Document Structuring and Retrieval Accuracy with ZeroEntropy

“ZeroEntropy’s reranker consistently outperformed every alternative we tested — recall@10, recall@5, across datasets. It’s fast, stable, and a no-brainer for legal document pipelines.”

— Malik Boudiaf, Co-founder at Equall

Company and Highlights

Equall is a frontier AI company building end-to-end systems to solve complex, real-world legal problems. Their products help lawyers and corporations evaluate and manage legal risk through automation and live data analysis.

Equall’s flagship platform processes thousands of VC and corporate documents — from certificates of incorporation to board consents — extracting structured information to help lawyers verify compliance and detect inconsistencies in real time.

With ZeroEntropy’s zerank-1 reranker, Equall achieved significantly higher recall and precision in their structured extraction workflows, leading to fewer false negatives and faster verification cycles.

• Domain: Legal document processing

• Use case: Structured extraction and deep document search

• Stack: Internal extraction logic + ZeroEntropy reranker

• Integration: Simple API swap

• Key metrics: Recall@10 ↑, Precision@10 ↑, latency <100 ms

Problem

In venture financing diligence, accuracy is non-negotiable.

Equall’s system parses hundreds of documents per deal, often concatenated PDFs with annexes, exhibits, and inconsistent formatting.

Their extraction pipeline needed to:

• Retrieve the right document segments for each target field (e.g., board resolutions, agreement dates).

• Maintain high recall across complex, domain-specific phrasing.

• Deliver consistent latency suitable for real-time lawyer workflows.

Off-the-shelf vector search or large-embedding-only systems struggled in this context — noisy retrievals, weak calibration, and insufficient context handling.

Approach

Equall integrated ZeroEntropy’s zerank-1 reranker on top of their existing retrieval pipeline, replacing a basic embedding-only system.

They benchmarked the reranker internally using annotated fields from real deal rooms, evaluating recall@k across multiple extraction types — from dates and grant approvals to board resolutions.

Implementation was frictionless:

1. Existing embedding candidates fed into ZeroEntropy’s API.

2. ZeroEntropy reranked the top candidates per field.

3. Minimal code changes, immediate lift in quality.

Results

Accuracy

• Consistent recall@5 and recall@10 gains across nearly all document types.

• Improved contextual understanding in ambiguous cases (e.g., multiple annexes or cross-referenced terms).

• Outperformed internal baselines and commercial rerankers tested.

Latency and Stability

• p50 latency: 75 ms

• p90 latency: 130 ms

• p99 latency: 240 ms

• Fully stable in production workloads with gradual rollout across extraction schemas.

Product Impact

• Reduced manual review time for lawyers.

• Increased trust in automated extraction.

• Enabled downstream “deep research” features powered by accurate retrieval.

Decision

Equall is now expanding ZeroEntropy’s use beyond reranking, testing ZeroEntropy newest models and end-to-end search pipelines.

“Our system relies on accurate retrieval — ZeroEntropy just works.”

— Pierre Colombo, Chief Scientist at Equall