Exploring the AI Legal Reviewer using OS RAG

https://opendatacircle.org/events/lakehouse-meetup-2

Transcript

1. Exploring the AI Legal Reviewer using OS RAG for Alec Lee, 2026/01/27 AX×EA Fullstack Explorer & Community Contributor - From Vision to Execution across Architectures, Cultures and Organizations. 🔍Exploration 🤝Co-Creation 🚀Empowerment Safety，Accuracy & Usability Purpose of Sharing: This is not about replacing legal staff, but about empowerment and making legal review verifiable, faster and responsible.
2. 「朋あり遠方より来たる、また楽しからずや。」 AX道を、共に歩み続ける仲間たちへ： To friends worldwide walking this AI Transformation journey together: "Is it not a joy when friends who share the same passion come from afar?" — from Confucius, The Analects -「孔子『論語』より」 https://chatgpt.com/
3. 1. Three Issues Analyzing current bottlenecks in AI legal review Scenario. 2. Open Source Architecting for safety and data & tech sovereignty. 3. RAG Accuracy Ensuring factual grounding and verified citations. 4. Visualized Workflow Usability for legal professionals without coding. 5. EA-Governed Roadmap The roadmap for successful implementation. Presentation Agenda
4. Q. “What does an AI Legal Reviewer actually change for a legal team?” As Is without AI: Contract Draft Receiving Analysing Terms Check list Comment on terms Co-worker double Check Contract Draft Sending To Be with AI: Contract Draft Receiving AI　 Analysing Check list AI Comment on terms PIC & Co-worker double Check Contract Draft Sending 70% of Lead Time Co-working between AI Team & Legal Staff Responsibility goes to Legal Staff A Part of Contract Lifecycle Management
5. Safety Data leakage risks when sending sensitive client contracts to public cloud LLM providers. Accuracy The risk of AI generating fake case law or fabricated legal citations that look convincing. Usability Difficulty for law staff without knowing coding to create workflow for different category of contracts. 1. Three Issues on AI Legal Review Current Solution: Deletion or Encryption of Company Name Customer Name Product Detail $ Amount etc Data Cleaning Prompt RAG under Blackbox of LLM/RAG GPTs / Gems … SaaS of Engineering of
6. 2. Open Source Architecture for Safety & Usability Unstructured Database AI Ready Intuitive Agent PF Chat UI Web/APP Tools (APIs & MCP) Models (Private) Architecture Design of AI Legal Reviewer https://github.com/zilliztech/attu Prompt Workflow Usability “Legal Team ”
7. PoC of AI Legal Reviewer Knowledge Embedding Prompt Engineering Model Installation Data Management
8. 3. RAG Setting for Accuracy Retrieval Testing on Dify Vector Search on Attu Setting of Chunk, Model, Retrieval Article on Knowledge Accuracy arises from the orchestration of controlled retrieval, persistent memory, LLM-driven agency and complementary components. Naive RAG Advanced RAG Multimodal RAG Graph RAG etc Ref:https://www.khuranaandkhurana.com/2023/07/24/non-disclosure-agreement-a-comprehensive-checklist-of-important-clauses-and-their-purpose
9. 4. Visualized Workflow Co-creation between AI Engineer & Legal Team members with visualized workflow
10. 5. EA-Governed Roadmap Transparency of Data Flow from data input to data output Knowledge Cleaning Chunks Vector DB Agent(LLM) Sustained ROI(Return on Investment) Role Reallocation of Shift from repetitive review to higher-value legal judgment KPI Value of Efficiency Improvement Investment of GPU and Maintenance
11. 「千日の稽古を鍛とし、万日の稽古を練とす 。」 "Forge with a thousand days, Refine with ten thousand days." – From Miyamoto Musashi, The Book of Five Rings AX道を、共に歩み続ける仲間たちへ： To friends worldwide walking this AI Transformation journey together: - 「宮本武蔵『五輪書』より」 https://chatgpt.com/
12. Thank you for your time & for the meaningful connection. Any questions or advice regarding the AI Legal Reviewer? 🔍Exploration 🤝Co-Creation 🚀Empowerment