Ngoda Boece

Paris · open to a 4 to 6-month internship in agentic AI, data or software development

Ngoda Boece — agentic AI & data

Student in AI & Data for Business at aivancity (Paris). I build AI agents and data tools, I build with AI agents, and I measure whether things actually work. 14 documented projects.

Agents I build

  • AQRS: 10 agents orchestrated with LangGraph in 13 steps, per-tool permissions, fallback between local models. LLMs never compute a number.
  • HelloTalk Assistant: a 100% local agent (Qwen via Ollama) that translates, corrects and suggests replies.
  • BLACKBOX: explores a web app, invents test scenarios, keeps only reproducible bugs (10/10 injected defects found).
  • PHANTOM: turns a task written in plain French into a tested, self-repairing web automation.

Agents I build with

  • Coding agents (Claude Code): I scope, specify, have code written, then review and test it.
  • BlenderMCP: an agent drives Blender through Python scripts. Two 3D short films, no imported 3D model.
  • Protocols before code: success criteria committed before the first line of the engine.

My guardrails

  • Nothing irreversible without human approval.
  • Automated tests, criteria fixed in advance, reproducible experiments.
  • Anything simulated is labelled as simulated.
  • Local first when data is sensitive.

Projects

Detailed pages are in French.