AI & Data - Applied AI Engineering: Multi-Provider APIs, Agents, and Connected Systems
Three hands-on days to integrate APIs from leading AI providers, use Claude Code and OpenAI Codex, build IoT-connected agents, and optimise reliability, token usage, and cost.
Detailed programme
Design a multi-provider AI architecture
- Compare models, modalities, context windows, latency, privacy, and cost for each use case
- Choose between direct APIs, enterprise cloud offerings, and a multi-provider abstraction layer
- Define routing, fallback, and portability strategies without unnecessary provider lock-in
Integrate leading AI API platforms
- Call the Anthropic Claude, OpenAI Responses, Google Gemini, and Mistral APIs through SDKs or HTTP
- Handle authentication, secrets, model selection, streaming, structured outputs, and tool calls
- Manage rate limits, errors, retries, timeouts, logging, and usage reporting
Multi-provider lab
- Implement the same service with two providers, then compare quality, latency, token usage, and fallback behaviour
Develop with Claude Code and OpenAI Codex
- Explore a repository, trace a flow, plan a change, and identify the relevant files
- Structure AGENTS.md, CLAUDE.md, progressive context, commands, and acceptance criteria
- Implement, test, and review a change while keeping the scope focused and human control explicit
Comparative lab
- Complete the same change with Claude Code and Codex, then compare the plan, modifications, and verification evidence
Build API-powered tools and agents
- Define structured tool contracts, execute tool calls, and return results to the model
- Implement an agent loop with state, memory, tools, stopping conditions, and human approval
- Connect tools and context sources through MCP with controlled read, write, and execution permissions
Design a robust agent harness
- Combine model, instructions, context, memory, tools, orchestration, guardrails, evaluations, and observability
- Define states, budgets, limits, stopping conditions, recovery paths, queues, and human approvals
- Plan for tool failures, ambiguous results, provider outages, and safe fallbacks before execution
Architecture workshop
- Design the harness for a production agent, including contracts, permissions, budgets, and expected evidence
Connect agents to IoT systems safely
- Expose device telemetry and commands through APIs, gateways, or MQTT brokers without giving the model direct access
- Separate read and action paths, then apply allowlists, approvals, timeouts, and operational limits
- Test with a digital twin or simulated device, log every action, and provide deterministic fallback behaviour
Supervised IoT lab
- Build an agent that analyses simulated telemetry and proposes or performs one strictly authorised action
Optimise context, tokens, and cost
- Measure input, output, reasoning, cached, and tool-use tokens where the provider exposes them
- Set a context budget using targeted retrieval, chunking, summarisation, and history compaction
- Reduce cost and latency with caching, model routing, bounded outputs, batching, and continuous measurement
Evaluate, secure, and observe agents
- Build representative scenarios and repeatable criteria for quality, security, and robustness
- Measure task success, latency, cost, human-intervention rate, tool calls, and validation coverage
- Integrate secret management, prompt-injection protection, traces, alerts, review, and feedback
Capstone and industrialisation plan
- Build a multi-provider agent with tools, an agent harness, and a simulated service or IoT-device connection
- Demonstrate the workflow, controls, results, token usage, and limitations
- Document the architecture, tests, operations, budgets, and next steps towards production
Tell us about your project
Our offices
- Exceev Consulting
61 Rue de Lyon
75012, Paris, France - Exceev Technology
332 Bd Brahim Roudani
20330, Casablanca, Morocco