I'm Erfouni, an AI/ML builder interested in systems that do more than generate textβthey retrieve useful context, call tools, automate workflows, validate outcomes, and produce real-world results.
- π§ Building with machine learning, RAG, and agentic workflows
- π Connecting models to tools and knowledge through MCP and REST APIs
- π³ Packaging ideas with Python and Docker
- βοΈ Exploring AI-assisted CAD and SolidWorks automation
- π§ͺ Current operating mode: prototype β test β break β learn β ship
| System | What makes it interesting |
|---|---|
| π§ AI knowledge systems | Retrieval pipelines that give models the right context before they answer |
| π Tool-using agents | MCP and API integrations that let AI take useful, controlled action |
| βοΈ Engineering automation | Intelligent workflows for CAD, design rules, and technical knowledge |
| π¦ Reproducible services | Python applications packaged to behave the same way outside my laptop |
I build OpenAI Codex plugins and MCP systems that connect AI-assisted SolidWorks workflows to curated engineering knowledge, deterministic validation, and consent-based learning.
design request
ββββΊ retrieve standards + known patterns
ββββΊ reason with constraints
ββββΊ call CAD tools
ββββΊ validate β learn β improve
| Project | Inspectable proof | Why it belongs here |
|---|---|---|
| SolidWorks GPT Plugin | A knowledge-backed Codex plugin with deterministic validation, install verification, and versioned releases | |
| Drawing API Database 2026 | 2,405 API members, 1,293 source-linked examples, 32 workflows, and bilingual FTS5 search in a validated SQLite artifact | |
| OpenRouter MCP Orchestrator | A tested MCP bridge for orchestrating multiple model families through one tool surface | |
| Helios Multi-LLM Orchestrator | A local-first MCP orchestrator with task decomposition, official-source benchmark routing, 22 unit and HTTP integration tests, and cross-platform setup |
No vanity counters hereβeach status links to code, tests, or a release you can inspect.
def build_idea(idea):
while not idea.works:
idea = debug(idea, coffee=True)
return dockerize(idea)
# "Works on my machine" is a bug report, not a deployment strategy.Waka waka β one contribution at a time.
Always learning. Occasionally debugging the debugger.