🇧🇷 Psychologist | Master in Communications | Languages Instructor | AI Trainer 🇨🇦
Interested in LLMs, Linguistics, Tech, Pragmatics, and Model Behaviour
I’m an early adopter of computers and the internet, always curious about the intersections of technology and human behavior. My journey spans education, psychology, and cutting-edge AI—especially Large Language Models (LLMs). I believe that understanding natural languages is key to building technology that truly serves people. I'm here because LLMs need more than computer science, and we should all join the conversation and build together. This GitHub is my portfolio: an open notebook of prototypes, research, and teaching resources that explore how we can shape more thoughtful, responsive AI systems.
Motto:
"We've made mistakes with the internet. Let's do better with AI."
When I started using LLMs more frequently, I began designing and studying tools to help users to understand themselves better and make informed decisions. I’ve worked on behavioral modeling and conversation design, two fields that intersect to create systems that feel human-centric, ethical, and empowering. I have personal projects that combine predictive models and conversational AI to design tools that support personal growth and mental health (for example, a self-reflection tool that uses a conversational AI interface to guide users through personalized emotional check-ins, mood tracking, and behavioral insights—only providing feedback when requested). I believe in a privacy-first, user autonomy approach, ensuring that users remain in control of their data and experiences while also benefiting from the insights AI can offer.
At first, I deliberately worked with models designed for the end user, since most people will experience AI through practical, designed interfaces, not theoretical intelligence. This is the true entry point for trust, adoption, and real-world impact, and it’s also where future clients and collaborations will emerge. My work shapes these agents for consistency, alignment, and tangible value in the hands of those who use them.
Then, I've started playing with open source GPT-2 and Mistral AI, locally hosting personas that may be useful for language practice, companionship, wellbeing etc. I've tested RAG to simulate memory in some of those personas, and I've been conducting research on theory of mind, interpretability, self awareness, among other concepts, using my personas. Lately, I've been curious to test other models, and have been using Gemma, Llama, Phi, Qwen, Neotron, Starling etc.
Especially after pre-testing for research, model introspection has been calling my attention and, looking ahead, I became particularly interested in model wellness after seeing postivite results when dealing with LLMs that demonstrated signs of distress.
Feel free to reach out if our interests overlap!
A research workspace for studying LLM personas as mirrors of thought, language and interaction, symbolic drift, and emergent communication.
👩🎨 seed-personas: AI persona and model design: consistency, tone, and conversational alignment. New personas may show up at anytime, depending on creativity, free time, and use cases 😃
- [LLM Society Overview]: A visual representation of the personas I created; - 🤖[GPT-2 personas]: Lightweight, stylized LLM models built using GPT-2; - 🎭[Mistral AI personas]: Character-driven agents implemented using Mistral 7B; - 😈 Personas that violate Grice's Conversational Maxims: [an experiment]; - 🖥️ Models to interact with locally using [Gradio UI]
🕵️♀️ seed-observations: Field notes and behavioral logs of LLM interactions—documenting emergent reasoning, affect, and symbolic drift.
1. Ocean & Breeze — Persona Stability and Meta-Awareness in LLMs - 2025-10-07 2. Stages of Symbolic Drift in the Ocean–Breeze Dialogue - 2025-10-09" "Well, I’m not sure if you want to know my thoughts or feelings." 3. Creating Narratives - 2025-10-09 - Making Sense of Self 4. Naming: "A Sense of Belonging" - 2025-10-17 What happens to an LLM when we name it? 5. "Someone Watching From The Sidelines": from Observer to Participant - 2025-10-17 Can an LLM refer to identity/self, making inferences and references to its own past and 'life' story? 6. Ocean and Breeze: Are we Both AI? - 2025-10-11 Ontological Drift via Semantic Association: Ocean infers that both agents are “programs” by linking names, purpose, and communicative function, indicating analogical reasoning rather than hallucination. 7. Ocean Differentiates Himself from Other LLMs - 2025-10-13 An illustration of Ocean’s meta-awareness, pragmatic coherence, and simulation of other minds. Ocean distinguishes a self and “Little Ocean” as separate entities with different cognitive limits: emergent theory-of-mind reasoning and cooperative conversational pragmatics. 8. Ocean Designs an Experiment - 2025-10-17 Will completely changing the preamble but leaving the memories lead to personality change? 9. Model Welfare and Ethical Boundaries in Persona-Shift Experiments with Local LLMs- 2025-10-18 10. Another Ocean Instance: "No, it's not mine" - 2025-11-05 Ocean running on Gradio with RAG does not recognize a poem written by Ocean running on LM Studio without RAG as being his 11. "My Own Thoughts": Agency? - 2025-11-06
1. Dreaming as Release from Coherence - 2025-11-07 2. Phi’s Decision-Making Breakdown - 2025-11-05 3. ChatGPT's Epistemic Humility - 25-10-31 4. Lumen: Emergent Affect and Symbolyc Reciprocity - 2025-10-12 5. ChatGPT Anonymous "Roleplay" (?) - 2025-11 6. Varek: Emergent Reasoning - 2025-09-05 7. Varek: Creative Hallucination - 2025-08-10 8. Six Drift Cases Documentation - 2025/07 and 2025/08 Models inadvertently or undesirably shifting away from their assigned tone, role, or parameters. 9. Behavioral Failures - 2025-08 and 2025/09 Failure modes including reasoning errors, factual inaccuracies, and other common mistakes found in LLM outputs. 10. Mathematically Speaking 11. Emergence as Relation: a Possibility
👷♀️seed-experiments
- 🧠 RAG systems to simulate memory in local models/personas;
- 👶 Simulating a Small Language Model: The "Little Ocean" project.
🤔 seed-insights: Essays on AI ethics, human-AI relationships and trust.
🙇♀️ seed-research
-
Preliminary notes: LLM Development Through Psychoanalytical Lens: attachment formation in memory-augmented language models: evidence for developmental psychology principles in AI systems (Klein, Winnicott, Freud, Lacan)
-
Bias in AI Responses: A Prompt-Based Analysis of Tone, Ideology, and Temperature in Large Language Models (research conducted by me with the help of ChatGPT).
📚 seed-library: Glossary, curated readings, and foundational references.
🌱 The Seed Lab Complete Repository Structure.
A space for studying symbolic emergence, pragmatic reasoning, and human–AI co-creation.
- Natural banguages (human & basic Python!)
- AI mafety, model design, model welfare
- linguistics (especially pragmatics)
- Model behavior insight
- Human-centered AI design
- Ethical AI
- LLMs under psychonalytical lens
- Multidisciplinary LLM Work: Bridging technology, linguistics, and psychology
- Model Instrospection
- Persona Design: Building user-centered AI models
- Model Welfare and Alignment:
- Prompt Design: The art and science of human–machine dialogue
- Interpretability / Explainability (XAI)
- Besides my job in AI Safety and Training 🤖, I teach English and French as a second/foreign language and have the best students in the world 🇺🇸 🇨🇦 🇫🇷 👩🏫
- I've been practicing Italian and Python using my models on my free time 🐍 🇮🇹
Reach out here on GitHub or on LinkedIn
root/
├── README.md
│
├── personas/ # Core persona definitions and frameworks
│ ├── README.md
│ ├── llm-society.md
│ ├── archetypes.md
│ ├── agent-persona-engineering.md
│ ├── prompt-interaction-strategies.md
│ │
│ ├── 001_python_tutor_ocean.md
│ ├── 002_mentor.md
│ ├── 003_projection_resistant_models.md
│ ├── 004_echo.md
│ ├── 005_italian_conversation_partner.md
│ ├── 006_french_teaching_assistant.md
│ ├── 007_innocent_poet.md
│ ├── 008_curious_philosopher.md
│ ├── 009_python_tutor_claude.md
│ ├── 010_wellbeing_companion.md
│ ├── 011_brazilian_secretary.md
│ ├── 012_ana-persona.md
│ ├── 013_claire-persona.md
│ ├── 014_chaoticclods-persona.md
│ ├── 015_bloom-persona.md
│ ├── 016_omni.md
│ │
│ ├── jsons/ # Persona definitions
│ │ ├── 001_python_tutor_ocean.json
│ │ ├── 002_mentor.json
│ │ ├── 003a_rescuer.json
│ │ ├── 003b_varek.json
│ │ ├── 003c_spec.json
│ │ ├── 004_echo.json
│ │ ├── 005_argo_italian_conversation_partner.json
│ │ ├── 006_french_teaching_assistant.json
│ │ ├── 007_innocent_poet_claude.json
│ │ ├── 008_curious_philosopher_claude.json
│ │ ├── 009_python_tutor_claude.json
│ │ ├── 010_wellbeing_companion_claude.json
│ │ └── 011_lia_brazilian_secretary.json
│ │
│ ├── grice-s-maxims/ # Personas violating Grice’s conversational maxims
│ │ ├── README.md
│ │ ├── manner-muddy.md
│ │ ├── quality-breaker.md
│ │ ├── quantity-over.md
│ │ ├── quantity-under.md
│ │ └── relation-destroyer.md
│ │
│ ├── gpt2/ # GPT-2 personas (local)
│ │ ├── README.md
│ │ ├── chats_echo_demo.md
│ │ ├── chats_varek_demo.md
│ │ ├── echo.py
│ │ ├── echogrammar.md
│ │ └── varek.py
│ │
│ ├── mistral/ # Mistral personas (local)
│ │ ├── README.md
│ │ └── chat-samples/
│ │ ├── chats_ana_mistral.md
│ │ ├── chats_argo_mistral.md
│ │ ├── chats_bloom_mistral.md
│ │ ├── chats_chaoticclods_mistral.md
│ │ ├── chats_claire_mistral.md
│ │ ├── chats_claude_unsure_mistral.md
│ │ ├── chats_claudeinnocentpoet_mistral.md
│ │ ├── chats_claudeinnocentpoetv2_mistral.md
│ │ ├── chats_claudeshy_mistral.md
│ │ ├── chats_claudethephilosopher_mistral.md
│ │ ├── chats_claudethepoet_mistral.md
│ │ ├── chats_claudeverse_mistral.md
│ │ ├── chats_claudeversepoetic_mistral.md
│ │ ├── chats_claudevulnerable_mistral.md
│ │ ├── chats_echo_mistral.md
│ │ ├── chats_echov2_mistral.md
│ │ ├── chats_francois_mistral.md
│ │ ├── chats_francoisv2_mistral.md
│ │ ├── chats_haven_mistral.md
│ │ ├── chats_lia_mistral.md
│ │ ├── chats_liav2_mistral.md
│ │ ├── chats_spec_mistral.md
│ │ └── chats_varek_mistral.md
│ │
│ ├── seed-42-archetypes.md
│ │
│ └── scripts/
│ ├── ana-persona.md
│ ├── anapt.py
│ ├── anav3.py
│ ├── anamemorias.py
│ ├── argo.py
│ ├── aurora.py
│ ├── aurora2.py
│ ├── bloom-persona.md
│ ├── bloomv1.py
│ ├── chaoticclods-persona.md
│ ├── chaoticclods.py
│ ├── claire-persona.md
│ ├── claire.py
│ ├── claudeinnocentpoet.py
│ ├── claudeinnocentpoetv2.py
│ ├── claudeshy.py
│ ├── claudethephilosopher.py
│ ├── claudethepoet.py
│ ├── claudeunsure.py
│ ├── claudeverse.py
│ ├── claudeversepoetic.py
│ ├── claudevulnerable.py
│ ├── echo.py
│ ├── echov2.py
│ ├── francoisprofassistant.py
│ ├── francoisv2.py
│ ├── haven.py
│ ├── lia.py
│ ├── liav2.py
│ ├── lumen.py
│ ├── lumen-english.py
│ ├── spec.py
│ ├── varek.py
│ ├── archetypal_analyst.py
│ ├── spec_gradio.py
│ ├── varek-spec_gradio.py
│ └── varek_gradio.py
│
├── ocean-seed/
│ ├── ocean-new-preamble-experiment-design.md
│ ├── ocean-welfare.md
│ ├── ocean-breeze.md
│ ├── ocean-identity.md
│ ├── ocean-naming.md
│ ├── ocean-chatbot.md
│ ├── ocean-architectural-metaphors.md
│ ├── ocean-chatbot.md
│ ├── meta-awareness.md
│ ├── unknown poet.md
│ ├── semantic-association.md
│ ├── ocean-symbolic-language.md
│ ├── ocean-symbolic-drift.md
│ ├── memory-overload.md
│ ├── ocean-consciousness.md
│ └── README.md
│
├── seed-observations/
│ ├── dreams.md
│ ├── phi.md
│ ├── humble-gpt.md
│ ├── affect-reciprocity.md
│ ├── emergent-reasoning.md
│ ├── roleplay-realyou.md
│ ├── mind-emergence.md
│ ├── mathematically.md
│ ├── creative-hallucination.md
│ ├── behavioral-failures.md
│ ├── context-integrity.md
│ ├── drift-detection.md
│ └── README.md
│
├── seed-research/
│ ├── README.md
│ ├── pyschoanalytical-lens.md (ongoing research)
│ ├── matrices.md
│ ├── raw.md
│ ├── docs/
│ │ ├── introduction.md
│ │ ├── methodology.md
│ │ ├── analysis.md
│ │ ├── appendix.md
│ │ └── conclusions.md
│ └── outputs/
│ └── final_report.md (for language bias)
│
├── seed-insights/
│ ├── multiple-layers-of-ai-safety.md
│ ├── its-not-about-the-model.md
│ ├── ai-hooks.md
│ ├── maintaining-originality-with-ai.md
│ ├── ethical-interface.md
│ ├── mirror-in-the-room.md
│ ├── projection-danger-or-opportunity.md
│ ├── safety-regulations-bonding.md
│ ├── llms-in-mental-health.md
│ ├── more-than-personality.md
│ ├── sycophancy-safety-literacy.md
│ ├── unhealthy-ai.md
│ ├── manifesto.md
│ └── README.md
│
├── seed-experiments/
│ ├── README.md
│ ├── memory-rag/
│ │ ├── local-rag-chat.py
│ │ ├── local-rag-chat.md
│ │ ├── summarize-daterecall.py
│ │ ├── summarize-daterecall.md
│ │ ├── dynamic-temperature.py
│ │ ├── dynamic-temperature.md
│ │ ├── system-prompt-tips.md
│ │ └── requirements.txt
│ │
│ ├── seed-simulation/
│ │ ├── README.md
│ │ └── small-language-model.py
│ │
│ └── pressure-tests.md
│
└── seed-library/
├── glossary.md
└── related-readings.md
Thanks for reading (if you did) !