AI Systems Engineer. I build systems around models — not prompts.
Most AI work stops at "which model should I use?". The harder question is what to build around it: memory, identity, tools, state, evaluation, and behavior that stays consistent over time.
Building THINK I.N.G — a cognitive systems lab — and documenting the whole process.
Three agents, built as entities with a voice rather than functions with a name:
Identity is not role. Steve is who and how it speaks; ChiefArchitect is what it does. Keeping those separate is what stops an agent from collapsing into a script.
Underneath it, a provenance-first knowledge wiki: raw sources in, contradiction detection, query and synthesis, relationship graph out.
Python stdlib only, Markdown on disk, no database until search actually hurts.
→ Vision, architecture & principles
Prototype ≠ validation. A system that runs is not an idea that is proven.
Cut half. Double the impact. Complexity has to earn its place.
Evidence before architecture. No framework until something real demands it.
Foundations over tools. Which fundamental survives when this technology disappears?
Python — systems, tooling, automation
TypeScript — interfaces, web
C++ — engines, performance
Unreal / Unity — gameplay, real-time
LLM agents & tool use
Multi-agent orchestration
Persistent memory & RAG
Real-time systems
AI Engineer + Gameplay Programmer, converging on Human–AI Interaction and real-time intelligent systems — systems that perceive, reason, remember, decide, act and interact.
Perceive ·
Reason ·
Remember ·
Decide ·
Act
Experiments, architecture decisions, dead ends and failures become the content.
Build → Experiment → Record → Explain.
Geovanni Soeiro — the systems, the process
Jogando-Lata — gameplays
Rio de Janeiro · 𝕏 @GeovanniSoeiro ·