Build systems
that
Cohort courses in AI Engineering, Agentic Coding, and System Design — for engineers who already ship, and want the judgement that separates a demo from something you can be paged for.
What's next
All times ISTEngineers from teams
that ship at scale
See what they built- Cohorts run
- 6
- Engineers taught
- 240+
- Projects shipped
- 180
- Average rating
- 4.8/5
- Company OneFintech
- Company TwoE-commerce
- Company ThreeSaaS
- Company FourHealthcare
- Company FiveLogistics
- Company SixGaming
Three tracks
Pick the gap you actually have
AI Engineering
Ship LLM systems that survive contact with real users.
Enrolling now
- Design a retrieval pipeline you can defend with numbers
- Build an eval harness before you build the feature
- Cut inference cost without quietly cutting quality
Agentic Coding
Build agents that hold up outside the happy path.
Starts 10 Nov
- Design tools an agent can actually use correctly
- Build execution loops that recover instead of spiralling
- Sandbox untrusted execution without crippling the agent
System Design
Design systems you could be paged for at 3am.
Cohort in progress
- Reason about consistency without reciting CAP
- Pick a datastore for stated reasons you can defend
- Design for backpressure before it designs for you
Honest fit
This is a bad fit for most people
Cohorts are small and the pace assumes you already write software for a living. Read both columns before you spend anything.
Come in if
- You have shipped and operated production software for a couple of years
- You can read an unfamiliar codebase without hand-holding
- You are the person your team asks when something breaks
- You want judgement, not a certificate
Skip this if
- You are learning to program — start elsewhere and come back
- You want a recorded course to watch at 2× on a commute
- You cannot protect five to six hours a week for two months
- You are looking for a job guarantee
How a cohort runs
Live, project-based, and small on purpose
01
Two live sessions a week
Ninety minutes each, 7:30–9:00 PM IST. Worked problems and code review, not lectures — the recording exists, but it is the consolation prize.
02
One cumulative project
You build a single real system across the cohort rather than a series of exercises. It ends up on the projects page with your name on it.
03
Small enough to be seen
Cohorts cap at 30–45 people. You will be asked to defend a design decision out loud at least once, which is the point.
04
Reviewed by people who ship
Written feedback on your project every week from engineers who operate systems like the one you are building.
The real comparison
You probably do not need a course
Most engineers get there on their own eventually. The question is what the eventually costs you.
Expensive in money and time. Cheapest in months lost.
- Pace
- Fixed. The cohort drags you forward whether you feel like it or not.
- Feedback
- Weekly, written, on your own code, from someone who has operated this.
- What you finish with
- A working system you designed, plus the arguments for why.
- Failure mode
- You fall behind, and it compounds. The dates are the whole mechanism.
Who teaches it
Pavan Jegurupati
InferLearn exists because the gap between a working demo and a system you can be paged for is mostly judgement — and judgement is hard to get from documentation. Every cohort is taught around systems that actually ran, including the ones that broke.
- Ships and operates the systems the courses are about
- Maintains open-source tooling used in production teams
- Writes and speaks about evaluation, agents, and distributed systems
From past cohorts
What changed afterwards
“I had shipped an LLM feature before, but I could never tell whether a change made it better. The eval module rearranged how our whole team works.”
“The tool-design sessions were the ones that stuck. Our agent stopped spiralling once we fixed the error surfaces rather than the prompt.”
“Starting from failure instead of the happy path is obvious in hindsight. I write design docs completely differently now.”
Built in the cohorts
Everyone leaves with something running
Backpressure Playground
An interactive simulator for queue backpressure - tune arrival rate, worker count, and timeout, then watch where the system collapses.
Retrieval Debugger
A trace viewer that shows why a RAG pipeline returned what it did - chunk scores, rerank deltas, and the query rewrite at each hop.
PR Triage Agent
An agent that reads an incoming pull request, runs the affected tests, and posts a review focused only on what the diff actually changed.
Writing
Notes from production
Reading a retry storm: the same incident in three graphs
The staging incident looked like a traffic spike on every dashboard we had. Here is what it actually looked like, and which graph would have told us in thirty seconds instead of twenty minutes.
Timeouts, retries, and the agent that took down staging
An agent with a three-second timeout and three retries turned one slow dependency into a thousand requests a minute. The fix was not a smaller number - it was a budget.
Build the eval harness before you build the feature
Teams that ship reliable LLM systems almost always wrote the measurement first. Here is what that looks like in practice, and why the ordering matters more than the tooling.
No. You need to be comfortable writing and operating software — a couple of years of real engineering. We teach the systems thinking around models, not the maths inside them.
Five to six hours a week: two live sessions plus project work. The project is cumulative, so falling behind compounds. If you cannot protect the time, take the next cohort instead.
Every session is recorded and available for the life of the cohort. The live hours are where the value is, though — the sessions are worked problems, not lectures.
Cohorts run on IST, typically 7:30–9:00 PM on two weeknights. Exact days are listed on each course page.
Yes. Full refund if you withdraw before the third session, no questions asked.
It will help, but that is a side effect. The course is organised around operating systems in production, which is a broader and harder target than the interview format.
Learn by
Shipping
Cohorts are small and they fill. If the timing is wrong, take the free workshop and come to the next one.