01
A working AI Operations Assistant
An agent that investigates a service incident using tools and returns structured analysis.
live, online, one real build
A 3-hour live engineering workshop. Build a working AI Operations Assistant that retrieves system information, calls Python tools, handles failures and returns structured answers.
Overview
Not an AI awareness session. You learn how an agent works internally, how tools are invoked, how execution is controlled and what happens when something fails.
An agent that investigates a service incident using tools and returns structured analysis.
Multiple tools, tool selection, multi-step execution and bounded iterations.
Exceptions, timeouts, invalid arguments, permissions and logging.
The project everyone builds
Developer asks: “Why is the payments API unhealthy? Check its status and relevant runbook.”
Reasons about the next tool call and whether more information is needed.
Deterministic status data.
Local knowledge lookup.
Observed evidence, possible causes, recommended next checks.
Tools use local mock data. No production system access and no autonomous remediation.
Hands-on labs
Lab 1
Define typed Python functions and expose their parameter schemas to the model. A model generating a tool call is not the same as executing it: your application validates arguments, runs authorised operations and returns results.
def get_service_health(service: str) -> dict:
return service_registry.get(
service, {"error": "Service not found"}
)Lab 2
Bounded loops, multiple tool calls, tool-call identifiers and why the full conversation state matters. Illustrative pseudocode; the repository has the complete runnable version with a pinned SDK.
for step in range(MAX_STEPS):
response = call_model(messages, tools)
if not response.tool_calls:
break
for call in response.tool_calls:
args = validate_arguments(call)
result = execute_allowed_tool(call.name, args)
messages = append_tool_result(messages, call, result)Lab 3
Deliberately trigger an invalid service name, an unavailable dependency and a tool timeout. Tell tool exceptions, model errors, empty responses and validation failures apart. Structured output enforces shape, not factual correctness.
Lab 4
Add a get_recent_incidents() tool and test whether the agent combines recent incidents, service health and runbook information into one answer.
Developer toolkit
Tech stack
Real, current tools used in production AI engineering. You install them once and use them in every session.
Python
Python 3.11+
OpenAI-compatible API
Any provider, free API routers work
Pydantic
Validate arguments and output
Pytest
Test failure scenarios
GitHub
Starter and completed repo
Agenda
LLM vs workflow vs agent, the execution loop, tool contracts and termination conditions.
Model API, request and response structures, instructions, context and token usage.
JSON Schema, tool definitions, argument validation, execution and tool results.
Multiple tools, tool selection, multi-step execution and bounded iterations.
Stretch, ask questions in chat.
Exceptions, timeouts, invalid arguments, idempotency, tool permissions and logging.
Extend the agent with a tool and test a failure scenario.
How MCP and LangGraph extend what you built.
Architecture questions, code resources and the full-cohort learning path.
3 hours including a 10-minute break. You code on a prepared repository with a controlled demo dataset, so time goes into the agent, not setup.
Outcomes
Code you wrote, ran and can extend.
Validation, permissions and results, not magic.
Timeouts, bad arguments and empty responses handled.
Clear next steps into MCP, LangGraph and the cohort.
Who it is for
A good fit if you
This may not be for you if you
Prerequisites
Required
Basic Python proficiency: functions, dictionaries, JSON, exception handling and fundamental API knowledge.
Recommended
async/await, type hints, REST APIs and Git. Prior LLM experience is useful but not mandatory. Students and recent graduates are welcome if they meet the programming prerequisites.
After the workshop
The workshop stands on its own. The same assistant can grow into a production-oriented prototype in the full cohort.
Instructor
Rajath Kumar K S
Founder of Analog Data. Teaches AI engineering through live online cohorts and workshops.
How registration works
Choose Register. It opens a dedicated secure page. You answer four short questions: background, Python level, LLM experience and what you want to solve.
Fee is ₹499 plus GST.
Confirmation, session link and the setup video arrive by email.
After the workshop
Tools, MCP, RAG, LangGraph, FastAPI, evaluation and deployment.
FAQs
No. It is for people who already write Python. If you are new to programming, start with Python basics first.
Python 3.11 or newer, VS Code, and a virtual environment or uv. A short setup video is sent before the session.
Bring any OpenAI-compatible API. Free API routers work too, so you do not need to pay for tokens to follow along.
Yes. The price is ₹499 plus GST, shown clearly before you pay.
No. You build one agent loop properly. MCP and LangGraph appear only as a preview. The cohort covers them in depth.
The curriculum is the same on each date. Pick whichever suits you and complete the project.
Yes. The workshop stands alone and shows how the same assistant grows in the 25-hour cohort.
Ready when you are
₹499 + GST · 3 hours live online · 28th Nov 2026, 6:00 PM to 9:00 PM IST
₹499 + GST
Register ↗