About
José Siqueira de Cerqueira is a Doctoral Researcher in GPT-Lab and the Gamification Group (Research Centre of Gameful Realities) at Tampere University, Finland. His research focuses on operationalizing AI ethics in software engineering through trustworthy LLM-based multi-agent systems and knowledge graph grounding, contributing to compliance with regulations such as the EU AI Act. His core research artifact, TERE4AI, generates legally traceable compliance requirements by combining a knowledge graph of EU AI Act provisions with a multi-agent requirements engineering pipeline. He is part of the Convergence of Humans and Machines programme funded by the Jane and Aatos Erkko Foundation, and contributes to the EVIL-AI project (also JAES-funded) and the SYNTHETICA project funded by the Research Council of Finland. His research interests include AI ethics, trustworthy AI, requirements engineering, software engineering, Large Language Models, and gamification. He holds a Master's degree in Informatics from the University of Brasília and has worked as a software developer for Brazilian government agencies.
Research Groups: GPT-Lab, Gamification Group (Research Centre of Gameful Realities)
Funded Projects
- Convergence of Humans and Machines (Jane and Aatos Erkko Foundation)
- EVIL-AI (Jane and Aatos Erkko Foundation)
- SYNTHETICA (Research Council of Finland)
In the Media
- YLE (Finnish national broadcaster): Tekoäly teki presidenttiehdokkaista karikatyyrit: testaa keskustelua virtuaaliehdokkaiden kanssa (January 24, 2024)
Acknowledged in
Acknowledged for designing and developing the Finnish presidential election chatbots · Finlandised electobots and the distortion of collective political memory (Cambridge University Press)
If you want to explore working together (workshops, joint research, or advisory on trustworthy AI), reach out: jose.siqueiradecerqueira@tuni.fi
Research
The problem. The EU AI Act is law, but it is written as 113 articles, 180 recitals, and 13 annexes of legal language. Organizations deploying AI must comply with it, and nothing in that text tells an engineering team what to actually build.
What I build. TERE4AI reads a plain description of an AI system and generates the compliance requirements it must meet, each one traceable to the exact provision of the Act it comes from. Under the hood: a knowledge graph of the entire Act linked to the EU's Trustworthy AI guidelines (590 semantic alignments), driven by a pipeline of four AI agents mirroring the requirements engineering phases.
Why trust it. A compliance tool built on LLMs has to earn trust in its own output. Every answer is grounded in the graph rather than the model's memory, a judge model audits the results, and coverage metrics show not just what the tool found but what it might have missed.
Ph.D. in Computer Science (in progress) · Faculty of Information Technology and Communication Sciences · supervised by Prof. Pekka Abrahamsson and Prof. Juho Hamari
For researchers: questions, method, and iterations
Research Questions
- RQ1: How can legislation and standards be structured into a knowledge graph to enable trustworthy LLM-based requirements generation?
- RQ2: To what extent does knowledge graph grounding improve the accuracy and coverage of LLM-generated compliance requirements compared to ungrounded approaches?
- RQ3: How does TERE4AI support appropriate trust calibration when developers use it for building EU AI Act compliant systems?
TERE4AI in Detail
TERE4AI (Trustworthy Ethical Requirements Engineering for AI) generates EU AI Act compliant requirements from natural language AI system descriptions. It combines a Neo4j knowledge graph with a multi-agent pipeline of four agents aligned to RE phases: elicitation, analysis, specification, and validation.
Knowledge Graph: 113 articles, 519 paragraphs, 375 points, 180 recitals, 13 annexes from the EU AI Act; 7 principles and 23 subtopics from AI HLEG; 590 semantic alignments between them.
Trustworthiness strategies: Multi-agent collaboration · Structured communication · Knowledge graph grounding · Self-assessment via LLM-as-judge · Coverage metrics · Source traceability
DSR Iterations
| Iteration 1: LLM-MAS | Iteration 2: RAG-Enhanced | Iteration 3: TERE4AI | Iteration 4: MCP Compliance Service | |
|---|---|---|---|---|
| Artifact | Multi-Agent System (no grounding) | RAG-Enhanced Multi-Agent System | Knowledge Graph + MCP + RE Pipeline | MCP-based Compliance Service |
| Exploration | EU AI Act Falls Short (ICSOB 2024) + PhD Research Plan (ICSOB 2024) | Lessons from Iteration 1: agents need access to legal documents | Mapping Trustworthiness in LLMs (arXiv, 2025): identified 20 strategies, KG among top 3 | Lessons from Iteration 3 evaluation |
| Prototyping | 3 agents (2 Python devs + 1 AI Ethicist), specialized roles, structured comm., 5 debate rounds, no legal doc access | Added RAG over EU AI Act, AI HLEG, ISO/IEC 42001 + web UI | Neo4j KG (590 alignments), MCP server, 4 RE-phase agents, LLM-as-judge, confidence indicators. Demo: prohibited + high-risk systems | (Planned) |
| Evaluation | 3 case studies (AI Incident DB), thematic analysis, hierarchical clustering, baseline comparison | Evaluated in master thesis (A. Khan, unpublished) | (Planned) Baseline comparison, expert validation, practitioner study | (Planned) |
| Communication | Can We Trust AI Agents? (TETHICS 2025) | Grounded Ethical AI (ICSOB 2024 demo) | TERE4AI Tool Paper (REFSQ 2026 Tools) + DS Paper (REFSQ 2026 DS) | (Planned) |
| Key Lesson | Multi-agent works but agents lack access to legal documents | RAG helps but legal text exceeds token limits, vector embeddings lose hierarchical structure | KG preserves legal structure; now expose as a service for external agents | - |
Publications
My publications, preprints, and datasets are openly available. Google Scholar · DBLP · ORCID · ArXiv · Zenodo
Talks & Workshops
Conference Presentations
- HEPP6, Sixth Helsinki Conference on Emotions, Populism, and Polarisation · The Political AI Chatbot Test: Convinced? · Helsinki, Finland · May 18, 2026
With Rebekah Rousi, Nuppu-Maija Pelevina and Henri Pullinen. Part of the political chatbot persuasion research line (EVIL-AI and SYNTHETICA). - CHASE 2026 / ICSE 2026 · Operationalizing AI Ethics in the Public Sector and Bridging Ethical Discourse · Rio de Janeiro, Brazil · April 2026
Presented both CHASE contributions; co-authored SEIP poster (EAI-USG) also presented. - REFSQ 2026 · Doctoral Symposium, Tool Demo, and Poster · Poznan, Poland · March 2026
Presented doctoral symposium paper, TERE4AI tool demo, and poster. Also acted as discussant. - TETHICS 2025 · Can We Trust AI Agents? · Vaasa, Finland · November 11-12, 2025
- ICSOB 2024 · Full Paper, Demo, PhD Retreat, and Best Poster Award · Utrecht, Netherlands · November 2024
4 contributions: full paper (Springer LNBIP), demo paper, PhD retreat paper, and poster presentation; won Best Poster Award. - Academic Mindtrek 2023 · Doctoral Consortium Poster · Tampere, Finland · October 2023
Exploring the Black-Box Problem and Ethical Issues in Gamification, Persuasive Systems and AI.
Invited Talks
- Master's student group, Tampere University · Trust in LLMs · Tampere, Finland · September 2, 2025
1-hour presentation invited by Olesia Prykhodko for a group of master's students. - ACAFE delegation (Santa Catarina rectors) · Trust in AI: LLM-based Multi-Agent Systems · Tampere, Finland · August 30, 2024
Presented to a delegation of 24 people including 12 rectors from the ACAFE system (Brazil). - University of Copenhagen · Trust in AI: LLM-based Multi-Agent Systems & Finnish AI Presidential Debate · Copenhagen, Denmark · April 25, 2024
Lunch talk hosted by Ville Vakkuri, funded by SYNTHETICA project. - Elisa Research Insights 2023 · LLM-based Multi-Agent System Demo · Helsinki, Finland · October 5, 2023
Poster and demo session at Elisa Research event. - University of Jyvaskyla Startup Lab · Are Conversational Generative AI Empowered Multi Robot Systems Ethically Aligned? · Jyvaskyla, Finland · November 17, 2023
Early PhD presentation on CGI multi-robot ethics and DSR methodology.
Workshops Given
- Agentic Programming series (3 of 3), GPT-Lab × Crazy Town · Advancing in the Harness: Beyond Basic Claude Code · Crazy Town, Tampere · June 8, 2026
MCP and how it meets skills and custom commands, AGENTS.md, planning agents in parallel, agentic memory, and building an AI-native application from scratch. - Agentic Programming series (2 of 3), GPT-Lab × Crazy Town · Beyond Vibe Coding: Exploring Agentic Development with Claude Code · Crazy Town, Tampere · April 29, 2026
From passive vibe coding to structured agentic development: sub-agents, custom skills, loops, planning mode, and parallel phased implementation. - Agentic Programming series (1 of 3), GPT-Lab × Crazy Town · Agentic AI in a Programmatic Way · Crazy Town Pendoliino, Tampere · January 30, 2026
From raw API calls to an agentic framework with Pydantic models and structured responses, ending in programmatic knowledge discovery. - Crazy Town Pendoliino Opening · Prompting Workshop · Crazy Town Pendoliino, Tampere · November 14, 2025
Fundamentals of prompting, advanced techniques (chain-of-thought, role-based, RAG), and hands-on session. - Tampere University · (Inter)disciplinary Perspectives to AI Ethics · Tampere, Finland · October 20, 2025
Presented "Mapping Trustworthiness in LLMs" at an interdisciplinary AI ethics workshop. - Tampere University (GPT-Lab Third Workshop AI) · Full Stack E-commerce with GenAI · Tampere, Finland · November 28, 2024
Students built and deployed a full stack e-commerce site using ChatGPT, v0, Cursor, FastAPI, SQLite, and Render.
Posters
- Bridging Theory and Practice: Ethical AI User Story Generator (EAI-USG) · ICSE 2026, SEIP track · Rio de Janeiro, Brazil · April 2026 (co-authored, with UnB and Tampere colleagues)
- TERE4AI: A Knowledge Graph-Based Tool for Generating EU AI Act Compliant Requirements · REFSQ 2026 · Poznan, Poland · March 2026 · paper
- Trustworthy LLMs for Ethically Aligned AI-based Systems: A PhD Research Plan · ICSOB 2024 · Utrecht, Netherlands · November 2024 · Best Poster Award · paper
- Exploring the Black-Box Problem and Ethical Issues in Gamification, Persuasive Systems and Artificial Intelligence · Academic Mindtrek 2023, Doctoral Consortium · Tampere, Finland · October 2023
Teaching & Supervision
At Tampere University
- Capstone Project Mentor, COMP.CS.530 Capstone project on LLM fine-tuning · Feb–Mar 2026
Mentored two independent student projects on LLM fine-tuning. - Guest Lecturer, COMP.CS.530 Fine-tuning Large Language Models · Jan–Mar 2025
Delivered a 45-minute lecture on mitigating hallucinations in LLMs ("Avoiding Hallucination: Ground Truth"). - Instructor, Full Stack E-commerce with GenAI · November 2024
Students built and deployed a full stack e-commerce site using ChatGPT, v0, Cursor, FastAPI, SQLite, and Render.
At Universidade de Brasília
- Workshop Instructor, Theory of the Consolidated Meta-analytic Approach (4h) · March 2021
- Workshop Instructor, Theory of the Consolidated Meta-analytic Approach (2h) · October 2020
- Workshop Instructor, LimeSurvey Questionnaire Creation Workshop (4h) · 2020
- Workshop Instructor, OBS Studio Video Recording Workshop (4h) · 2020
- Teaching Monitor, Computação Básica (Basic Computing) · 2010–2011
Three semesters as paid tutor/monitor.
Supervised Theses
| Student | Title | Status | Year |
|---|---|---|---|
| Ayman Asad Khan | Can Software Development Adopt AI Ethics? Agents4EthicalSE: A Multi-Agent RAG Framework | completed | 2025 |
| Muhammad Kamran Khan | Validating AI-Generated Content: Security Challenges and Solutions Inspired by OWASP | completed | 2026 |
| Shayla Parvin | Mitigation Strategies for the OWASP Top 10 for LLM Application 2025: A Systematic Literature Review | completed | 2026 |
| Md Sadidur Rahman | Implementation and Evaluation of OWASP LLM Security Mitigation in Multi-Agent Systems: A Design Science Approach | completed | 2026 |
Examiner for 7 master's theses
Tools & Code
Research Tools
- TERE4AI · Knowledge graph + multi-agent pipeline for EU AI Act compliant requirements
- RE4AI Ethical Guide · Interactive guide for ethical requirements elicitation in AI · live tool
- Evil-AI Benchmark · Adversarial security benchmark for LLM agents in smart environments (IEEE Access 2026) (contributor)
- Tone Navigator · Multi-agent pipeline analyzing CEO tone in earnings calls (industry collaboration, code not public)
Side Projects
- zotero-watch-folder · Zotero plugin that watches a folder and auto-imports whatever lands in it
- workshop-CrazyTown-1 · Agentic AI in a Programmatic Way: from raw API calls to agentic frameworks
- workshop-CrazyTown-2 · Beyond Vibe Coding: structured agentic development with Claude Code
- workshop-CrazyTown-3 · Advancing in the Harness: MCP, skills, planning agents, AI-native apps
Datasets: 7 datasets on Zenodo
Awards & Service
Awards
- Best Poster Award · ICSOB 2024, Utrecht, Netherlands (2024)
- Honourable Mention · University of Brasilia Dissertation Prize (2022)
- 1st Place · Software Engineering Doctoral and Master Thesis Competition, SBES (2022)
- 3rd Place · Thesis and Dissertations Competition, SBSI (2022)
- Best Paper Award · IADIS International Journal on www/Internet
Organizing
- Organizing team member, GAISE 2026, 2nd Summer School on Generative AI in Software Engineering · Tampere University, June 1-3, 2026, hosted by GPT-Lab
- Co-organizer, AI-SW-Hackathon 2024 · AI & Software in Agriculture, Tampere University
- Co-organizer, AI-SW-Hackathon 2023 · AI & Software challenges, Tampere University
Reviewing
- Information and Software Technology (IST) (6 reviews, 2025 to 2026)
- ACM Transactions on Software Engineering and Methodology (TOSEM) (2025)
- Cognitive Systems Research (2026)
- Social Sciences & Humanities Open (2025)
- HICSS (2021, 2022, 2023, 2026)
- TETHICS 2025
- SOICT 2024
Background
Education
- Ph.D. in Computer Science (in progress) · Tampere University, Finland (2023–2027)
- M.Sc. in Informatics · Universidade de Brasília, Brazil (2021)
- B.Sc. in Computer Science · Universidade de Brasília, Brazil (2017)
- Specialisation in IS Management · Faculdade JK, Brazil (2020)
Exchange Programs
- Universitat Politècnica de Catalunya (UPC), Barcelona, Spain (2013–2014) · CNPq scholarship
- Universidad Complutense de Madrid (UCM), Madrid, Spain (2012)
Work Experience
- Doctoral Researcher · Tampere University (2023–present)
- Software Developer / Researcher · Brazilian Intelligence Agency (ABIN) (2019–2024)
- Software Developer · SERPRO (Federal Data Processing Service) (2021–2023)
- Researcher · Institut de Recherche pour le Développement (IRD), France (2021–2022)
- Researcher · Ministry of Justice of Brazil (2020–2022)