Custom AI assistants with curated legal knowledge bases, queried in de-identified form¶
Design · All designs
Evidence · no claims cited
Description¶
The method's operational key tool in both units was a custom AI built on free-tier platform functionality (Projects in ChatGPT and Claude, Gems in Gemini) loaded with the unit's applicable normative base. As the article describes it, "By loading into the custom AI the normative base and the correct documents pertinent to the unit’s subject matter, the public servant obtains answers anchored in that base, with persona and tone configured for the institutional context." At SES/CONT the base comprised ten legal instruments (decrees, normative instructions, laws, manuals); at UCI/SEDET, modular custom AIs were built per recurring subject (e.g., Public Procurement under Law No. 14.133/2021). Queries were made de-identified, under a Data Anonymization Protocol, so submitted content ceases to be personal data under Article 12 of the Brazilian Data Protection Law.
Design Implications¶
Context¶
Requirements¶
- A curated documentary/normative knowledge base pertinent to the unit's subject matter, loaded into the custom AI before use.
Constraints¶
- Queries must be de-identified: no names, case numbers, or identifiable data of the concrete case enter the tool; conversation-training features are deactivated and temporary conversation modes used.
Target Learners¶
- Public servants producing legal and technical administrative documents in internal control units
Learning Goals¶
- Drafting legally grounded documents, answering normative questions, and standardizing technical writing without exposing protected data
Claims¶
- No Ai Incidents Under Human In The Loop Policy [+M]
Related Designs¶
- Seven unit-specific custom GPT chatbots for pharmaceutical science units
- Hybrid query-resolution platform routing queries across AI assistant, peer layer, curated FAQ corpus and administrator escalation
Examples¶
-
Key Sources¶
- Vinicius Santana Gomes. (2026). The Main Barrier to AI Adoption in the Public Sector Is Lack of Training: How a Structured Method Accompanied Productivity Gains in Two Brazilian Government Cases. arXiv. https://arxiv.org/abs/2606.01517
Page metadata
| Field | Value |
|---|---|
type |
design |
id |
custom-ai-legal-knowledge-base-assistant |
title |
Custom AI assistants with curated legal knowledge bases, queried in de-identified form |
description |
The method's operational key tool in both units was a custom AI built on free-tier platform functionality (Projects in ChatGPT and Claude, Gems in Gemini) loaded with the unit's applicable normative base. |
status |
review |
generated |
by process:settle-candidates on 2026-10-10 |
trust |
unverified — no one has confirmed this page's content |
sources |
1 — vinicius-santana-gomes-2026 |
Frontmatter is the page's machine-readable record — see the Schema & Guide.