Symio-ai/legal-deposition-analyzer

Model Description

Legal Deposition Analyzer processes deposition transcripts to extract: key admissions, contradictions, impeachment material, factual timelines, evidence of spoliation, and testimony that supports or undermines each cause of action element. It also identifies inconsistencies between depositions and other discovery materials.

Transforms raw transcript text into structured litigation intelligence.

Intended Use

• Primary: Analyze deposition transcripts for impeachment material and key admissions
• Secondary: Cross-reference testimony against document discovery for contradictions
• Integration: Discovery analysis tool in the GLACIER pipeline

Task Type

text-generation -- Structured extraction and analysis from deposition text

Base Model

meta-llama/Llama-3.1-8B-Instruct -- Long-context instruction-following for transcript analysis

Training Data

Source	Records	Description
Deposition Transcripts	~100K	Full deposition transcripts (anonymized)
Impeachment Examples	~50K excerpts	Identified contradictions used at trial
Trial Cross-Examination	~30K transcripts	Cross-examination using deposition testimony
Expert-Annotated Transcripts	~10K	Attorney-annotated key passages
Document-Deposition Pairs	~200K	Documents referenced in depositions

Analysis Output Format

DEPOSITION ANALYSIS: [Deponent Name] - [Date]

KEY ADMISSIONS:
1. Page X, Lines Y-Z: "[quote]"
   Supports: [which element of which cause of action]
   Impeachment value: HIGH

CONTRADICTIONS:
1. Page X vs. Document [Exhibit #]:
   Testimony: "[quote]"
   Document says: "[quote]"
   Impeachment strategy: [how to use at trial]

TIMELINE EXTRACTED:
  [date]: [event] (page X, line Y)

ELEMENT SUPPORT MATRIX:
  [For each COA element, which testimony supports/undermines]

CREDIBILITY ASSESSMENT:
  Evasive answers: [count and examples]
  "I don't recall" responses: [count]
  Contradicted by documents: [count]

RECOMMENDED FOLLOW-UP:
  1. [Additional discovery needed based on testimony gaps]

Benchmark Criteria (90%+ Target)

Metric	Target	Description
Key Admission Recall	>= 88%	Identify 88%+ of important admissions
Contradiction Detection	>= 85%	Find contradictions between testimony and documents
Timeline Accuracy	>= 92%	Extract correct dates and events
Element Mapping	>= 85%	Correctly map testimony to COA elements
Impeachment Utility	>= 80%	Identified material is actually useful for impeachment

GLACIER Pipeline Integration

Post-discovery phase:
  deposition-analyzer processes all transcripts
  Output feeds into: element-mapper, timeline-builder, evidence-classifier
  WDC panels use analysis for trial preparation strategy

Training Configuration

• LoRA fine-tuning (r=16, alpha=32)
• Epochs: 3
• Learning rate: 1e-4
• Max sequence length: 16384 (depositions are lengthy)
• Hardware: AWS SageMaker ml.g5.12xlarge

Limitations

• Deposition transcript quality varies (OCR errors in older transcripts)
• Cannot assess witness demeanor or credibility from text alone
• Video deposition analysis is not supported (text-only)
• Complex technical testimony may require domain-specific knowledge
• Does not generate deposition questions (that is a separate tool)

Version History

Version	Date	Notes
v0.1	2026-04-10	Initial model card, repo created