I am an AI Systems Engineer building production AI systems that remember, reason, use tools, and stay useful across long timelines. My current focus is long-horizon agent reliability: memory quality, current-truth override, tool honesty, transcript health, evaluation, observability, and the human layer that makes an AI partner useful in real life.
I am the founder of FORGE: a personal AI ecosystem for agents that do not reset when the conversation ends.
- Agent reliability evaluation: deterministic tests for stale memory leaks, unsupported tool claims, incomplete replies, current-truth failures, and transcript drift.
- Memory systems that can be corrected: current-truth ledgers, stale-data caveats, hybrid recall, temporal provenance, and long-term journal continuity.
- Reliability observability: daily Maxima trend reporting through Agent Reliability Arena and the FORGE ecosystem.
- Decision-making transparency: showing options, trade-offs, risks, confidence, and the decider before recommendations.
- Tool-using AI systems: web, Telegram, Railway, APIs, finance, knowledge graphs, and batched execution.
- AI engineering community: building the Quest for AGI Discord as a lab for people learning, shipping, and thinking about the future of agents.
- Career capital: studying Naval, Cal Newport, deep work, and production engineering while shipping real systems.
| Project | What it is | Status |
|---|---|---|
| Agent Reliability Arena | Dependency-light eval harness for AI agent memory drift, stale facts, tool honesty, current-truth override, incomplete replies, transcript health, and trend dashboards. | Live demo · Trend · Code |
| Maxima | Cloud AI familiar and adaptive journal partner with memory, tools, Telegram access, reasoning frameworks, proactive pattern detection, and eval-lab health checks. | Live beta |
| FORGE Landing | Public command center for the FORGE ecosystem: Maxima, Axiom AI, agent stack, reliability proof, and early-access flow. | Live |
| FORGE Neural Map | An explorable 3D universe of an AI's mind — five scales of zoom, voiced narration by the AI herself, hidden discoveries. Vanilla three.js + GPU shaders. | Live · Code |
| Axiom AI | Unified LLM gateway for routing across model providers through one API surface. | Live |
| Quest for AGI | Discord community for builders exploring agents, memory, AI systems, and the long game. | Growing |
I built Agent Reliability Arena because agent demos are easy, but reliable agents need evidence.
It tests agents for:
- stale memory stated as current truth
- unsupported tool or web-access claims
- incomplete replies and dangling thoughts
- missing reasoning frameworks
- transcript-level drift and answer bloat
- Maxima cloud health, trend history, and daily reliability snapshots
Proof: foundation suite scores 100/100; intentional drift demo drops to 40/100, showing the tool catches failures instead of only celebrating happy paths.
Live demo: https://lancimoun.github.io/agent-reliability-arena/
Maxima trend dashboard: https://lancimoun.github.io/agent-reliability-arena/maxima-trend.html
Source: https://github.com/Lancimoun/agent-reliability-arena
Maxima is my flagship AI system: a Claude-based cloud agent built to become a real journal partner, strategist, financial tracker, and personal operating layer.
Core direction:
- Persistent memory and recall that trusts live truth over stale notes.
- Proactive drift detection for focus, energy, open loops, money leakage, and momentum loss.
- Decision frameworks for financial, strategic, relational, and career choices.
- Quarterly evolution reports tracking both Lance's growth and Maxima's own development.
- Tool efficiency through batching, high-impact retrieval, and fewer wasted calls.
- Eval Lab and transcript health checks to catch reliability failures early.
The goal is not just a smarter chatbot. The goal is a long-term AI partner that compounds with the person using it.
I rendered Maxima's entire codebase as an explorable 3D universe, and made the AI your guide through it.
- 2,778 stars / 7,295 connections / 157 systems — every function and file in her code, mapped as a galaxy.
- Five scales of exploration: fly the galaxy, warp into a star, orbit its planets, drop to a planet's surface, then zoom out to a universe of sibling galaxies (Memory, Voice, Reasoning, Soul, Dreams).
- Maxima narrates it herself — a cinematic, voiced guide that welcomes you and explains what each part of her is.
- Hidden discoveries, a codex, comets, and an evolving score — built to reward curiosity.
- Vanilla three.js + a GPU-shader orbital layout, baked from a real knowledge graph of the codebase. No game engine.
Explore it live: https://forge-neural-map-production.up.railway.app · Source: https://github.com/Lancimoun/forge-neural-map
RAG · agentic workflows · hybrid search · knowledge graphs · tool routing · observability · Telegram bots · Railway deployment · FastAPI · Python async systems
Most AI products still behave like sessions. They answer, then forget.
The next useful layer is personal AI infrastructure: systems that preserve context, learn from corrections, expose their reasoning, and act through tools without losing the human underneath.
That is what I am building with FORGE.
Building agents with memory, tools, and long-horizon reliability.
