Legal AI Engineering Training Course

Day 1: Building the Foundation — Ingest, Search, Retrieve

Module 1: The Legal Engineer’s Landscape

• Learning objectives—understand the role, where AI fits in legal work, and the two critical risks that permeate everything.
• Topics
• • The legal-engineer role and current market demand
  • Where AI fits: eDiscovery, review, contracts, research, investigations; the EDRM model explained simply
  • Build vs. buy considerations
  • The two core risks: confidentiality/privilege and defensibility

Module 2: Legal Data Is Messy — Ingestion and Extraction

• Learning objectives—handle the reality of legal data at scale.
• Topics
  • 1,400+ file types, email and PST formats, scanned paper, load files (.dat/.opt); critical embedded metadata
  • Text extraction (Tika), OCR, and deduplication as strategic choices
• Lab: FreeEed Ingestion—build an ingestion pipeline over a deliberately messy document set (email/PST, scans, load files)

Module 3: Search and Retrieval — The Foundation

• Learning objectives—construct the core eDiscovery primitive: finding anything inside everything.
• Topics—full-text search and indexing (Solr/Lucene); relevance, metadata and date filtering; searching across OCR’d content
• Lab: eDiscovery Search—index a corpus and run real eDiscovery-style searches, including within OCR’d scans

Module 4: RAG for Legal Documents — With Citations

• Learning objectives—build RAG over legal documents that cites its sources.
• Topics
  • Why retrieval, not fine-tuning, is preferred for sensitive material—the model never consumes the documents directly
  • Chunking, embeddings, and crucially citations/provenance
  • Multi-document and thread summarization
• Lab: Legal RAG with Citations—build a RAG Q&A over a document set that answers with source citations

Day 2: Making It Private, Defensible, and Shippable

Module 5: Privacy, Privilege, and Local Serving — The Privilege Trap

• Learning objectives—keep legal data local and certifiable.
• Topics
  • Where data goes when it reaches a cloud AI
  • Privilege waiver, duty of competence, and the “private” spectrum (contractual vs. physical)
  • Morgan v. V2X case study and why local deployment is court-defensible
  • Serving local models (Ollama/vLLM) and monitoring outbound traffic
• Lab: Local Model + Egress Proof—run a local model end-to-end and prove, via monitoring, that no data left the premises

Module 6: Defensible AI Review

• Learning objectives—measure and document an AI review so it withstands legal scrutiny.
• Topics
  • Court-worthy metrics: recall, elusion, precision, ground-truth validation; TAR/active learning
  • Transparency (why did it code this document?) and reproducibility—pin the model, fix settings, log everything
  • The “defensible case snapshot” enabling someone to re-run your review a year later with identical results
• Lab: Defensible Review—measure an AI review against a blind ground truth and produce a reproducibility bundle

Module 7: Ship It — Workflow, Private Deployment, and Governance

• Learning objectives—assemble components into a workflow, deploy privately, and score the system.
• Topics
  • A multi-step legal workflow (ingest → search → summarize → review → produce) with human-in-the-loop
  • Private/on-premises deployment essentials (containerization; keeping data in-house)
  • AI governance for legal in brief, and scoring the system using SAIS-100 (the Elephant Scale Secure AI Score)
• Lab: Score and Package—wire a multi-step workflow, score it with SAIS-100, and package it for private deployment

Capstone (integrated across Day 2)

• Build a private, defensible legal-AI application end to end—ingest a messy corpus, search it, answer questions with citations using a local model, measure a defensible review, and package for private deployment.
• Participants leave with a portfolio project that mirrors the legal-engineer job role.

Optional Day 3 / Advanced Modules (deliverable as a 3rd day or a modular series)

• Investigations: Entities, Relationships, and Timelines—extract people/orgs/dates, reconstruct email threads, build chronologies, map near-duplicates and document lineage. Lab: build a timeline and entity/relationship view.
• Agentic and Multi-Step Legal Workflows (deep)—richer orchestration, contract analysis, multi-doc synthesis, tool use, and guardrails as a design principle. Lab: build a multi-step workflow with a human checkpoint.
• Deployment at Scale—on-premises and appliance deployment, distributed processing for large volumes, regulated environments (CJIS, government, higher-ed), hardware sizing. Lab: containerize and scale a processing job across workers.
• Governance and Compliance Deep-Dive—the AI-regulation landscape (100+ US state AI laws, the EU AI Act), audit requirements, and a full SAIS-100 governance audit. Lab: audit a legal-AI system against a governance/defensibility checklist.