Orchestra is the operational and technical arm of Stratelya’s AI practice. We analyse a process, simplify it, connect your tools and put AI agents into service, with human sign-off, monitoring and measurement. The same building blocks run our own agents.
More than twenty agents run on our own processes today. Pick a domain: what each agent does, what triggers it, what stays human.
Reads over 500 news feeds, the global GDELT database and official registries, then matches each new publication against the companies followed.
Sends the alert by e-mail, Slack or Microsoft Teams, with its source and date.
Searches the last 24 hours for developments with business impact: sanctions, tariffs, conflicts, AI and data regulation, cyber incidents. Each item is tied to its source.
Scores posts from 1 to 10 for relevance and surfaces those that mention the companies followed.
Collects French and European notices, qualifies them with a fixed grid and no AI, then analyses the notice and tender documents, citing pages.
Extracts the details of a tender received by e-mail, as a web page or as a document.
Each step produces reusable technical artefacts and ends with a decision point backed by your own measurements.
Measure the real process before choosing where an agent goes.
For each layer: what we use in service and the market building blocks we are evaluating. The final choice is made at diagnostic, to fit your constraints.
Every run is traced and reviewable.
Solid line: in service at Stratelya. Dashed: market building blocks under evaluation, not used yet.
At a client, we reuse these building blocks or connect to its tools through their APIs. The choice of models, orchestrator and hosting is made at diagnostic.
A useful agent is an engineering problem first, not a prompt. These are the patterns we apply, the problem each one solves and its status with us.
We set the autonomy of each task on two criteria: the cost of an error and whether it can be undone.
The agent analyses and recommends; a person decides and acts.
Human decision and execution.
The risks specific to language-model applications, per the OWASP Top 10 for LLM Applications, and the countermeasure we apply.
Compliance: each use is classified under the AI Act from the diagnostic; personal data processing is minimised and retention periods set, in line with the GDPR.
Pick a function to see a concrete example: the process before and after, the tasks handed to agents, the responsibilities that stay human, the connected tools and the deliverables.
Tool categories: actual connections depend on your environment and are confirmed during the diagnostic.
These examples illustrate the approach. They are neither client references nor achieved results: each process is analysed from how your company really works.
A complete journey: a client need comes in, agents prepare, your teams sign off. No real events are published for this journey: this walkthrough is a simulation.
Start the simulation to follow the journey step by step.
Simulation: no client data, no real events. Agent names and contents are examples.
So you know exactly what you are buying, we separate what Stratelya delivers today, what this page illustrates and what gets built for you.
They illustrate the approach: they are neither client references nor results.
Every agent has an owner in your teams, a written scope and operating rules.
Prices, commitments, choices about people, payments: the agent prepares, a person decides.
Access limited to what is strictly needed, agent actions logged and reviewable.
Nothing is rolled out more widely without measurement on the pilot and your decision.
A first conversation to find the process where effort is lost the most. We start from how your company really works, not from a sales demo.