R&D prototypeLangGraph · MLflow · HETIC
Multi-Agent Customer Support
Seven LangGraph agents for intake, classification, sentiment, intent, prioritisation, routing and synthesis — tracked with MLflow.
Narrow agents, explicit graph, measurable runs.
01Context
Master's project at HETIC, Paris.
02Problem
Triage support messages end-to-end with specialised agents rather than one prompt.
03Constraints
- Academic timeline
- Explainable routing
04My role
What I personally built
- Agent decomposition into seven responsibilities
- LangGraph orchestration
- MLflow run tracking
What others owned
Independent project — no team. Everything here is mine.
05Architecture
Architecture explorer
The business flow — what happens, in plain words.
01 / 04
Message in — A customer writes.
06System
A LangGraph state graph passes each ticket through intake → classification → sentiment → intent → priority → routing → synthesis.
07Challenges
- Keeping shared state consistent between agents
08Outcome
Working multi-agent pipeline with tracked experiments.
09What I learned
Tracking agent runs like ML experiments makes prompt changes comparable.
10Stack
- LangGraph
- Python
- MLflow
- LLM
11Evidence
- HETIC Master's project, 2026