// Discipline 02
Agents, MLOps and observability
From agent architecture to the model in production: governed agents, measurable evaluation and observability to know what the system does and why.
01 · Who it is for
With a prototype that works in the demo and fails on real data.
That need architecture, evaluation or MLOps for agents and LLMs.
That must be able to explain, audit and limit what a model does.
02 · How we work
Agent architecture, model choice and an evaluation plan before building.
Agents, RAG and inference pipelines with traces, limits and tests, also on your own infrastructure.
Test suites, prompt red teaming and error measurement in production.
The open course QUANT AGENTIC ORCHESTRATION and custom workshops for teams.
03 · Specialties
04 · Method
What data exists, which decision depends on it and what has failed so far.
Hypothesis, method and success criteria in writing, before building anything.
Reproducible code and analysis, tested against adverse cases.
Everything is documented and in your team’s hands.
05 · Sectors
06 · Use cases
Illustrative cases: they describe the kind of work we do, not specific clients.
07 · Data and topology
An agent system is a graph: tools, memories, models and people connected. We design and measure it as such, node by node and path by path.
08 · Research
Second entry in the System 7 field diary — our poker agent for dev.fun's Arena: the pivot to heads-up, the node-locking that finally exploits for real, the v1.5 survival engine, and the war stories of running it in production.
System 7A technical field report on building a poker agent for dev.fun's Arena: the heuristic method, LLMs at the table, and the honest results.
System 7Tell us the context: what data exists, which decision depends on it and what has failed so far.