AI Engineering — RAG, Agents & Production LLM Systems
Build AI systems that survive real users, real costs and real evaluation.
A three-month live course for developers and analysts who want to build AI systems rather than use them: retrieval-augmented generation end to end, evaluation and guardrails, agents with bounded tool use, and the deployment, observability and cost control that separate a working system from a demo. The capstone is a deployed RAG assistant with a regression suite in CI.
- Duration
- 3 months
- Live teaching
- ~50.5 hours
- Fee
- ₹14,999
- Format
- Live online
Who this is for
Developers, analysts and technical founders who can already write Python and want to ship AI systems people depend on.
Prerequisites
- Comfortable writing Python (functions, classes, packages)
- Comfortable with the command line and Git
- Able to read API documentation without hand-holding
Who it is not for
Complete beginners. If you have not written code before, take Generative AI & Vibe Coding first — this course will not slow down for the basics.
Compare all five courses →Outcomes
What you will be able to do
Written as capabilities rather than topics, because a topic list tells you what was said in the room and not what you can do afterwards.
- Build a retrieval pipeline that returns the right context
- Ship RAG with citations over documents that keep changing
- Evaluate an AI system with metrics rather than impressions
- Build agents with bounded tool use and real stopping conditions
- Deploy, monitor and cost-control an LLM application in production
- Defend your architecture decisions in a technical interview
Curriculum
8 modules, ~50.5 live hours
Every module states the hours it takes and what you can do at the end of it.
01Python for AI engineering
6 hrs
Python for AI engineering
- Async Python and why it matters when every call takes two seconds
- Type hints, Pydantic and validation at the boundary
- FastAPI: routing, dependencies, background work
- Testing code that calls a non-deterministic API
After this module: A tested API service you can deploy and change without fear.
02LLM internals that actually matter
5 hrs
LLM internals that actually matter
- Tokens, context windows and what gets silently truncated
- Temperature, top-p and determinism in production
- Latency budgets, streaming and perceived speed
- Cost modelling per request, per user, per month
After this module: Predict what a feature will cost per thousand calls before building it.
03Structured output and tool calling
5 hrs
Structured output and tool calling
- JSON schema, function calling and enforcing a contract
- Validation, retries and graceful degradation
- Fallback chains across models and providers
- Making model output safe to write to a database
After this module: Turn a model’s answer into data your system can trust.
04Embeddings and vector search
6 hrs
Embeddings and vector search
- What embeddings encode, and what they do not
- Chunking strategies and the retrieval failures each one causes
- Vector, keyword and hybrid search
- Reranking, and why it usually matters more than the embedding model
After this module: Retrieve the passage that answers the question, not one that resembles it.
05RAG end to end
8.5 hrs
RAG end to end
- Ingestion: parsing, cleaning, metadata, incremental refresh
- Indexing and keeping an index current as documents change
- Retrieval, context assembly and prompt construction
- Citations that point at the actual source span
- The failure modes: missing context, conflicting sources, stale answers
After this module: A working RAG assistant over private documents, with real citations.
06Evaluation and guardrails
6 hrs
Evaluation and guardrails
- Building a golden set before you need one
- Retrieval metrics and answer metrics, measured separately
- Hallucination and groundedness checks
- PII and prompt-injection defences
- A regression suite that runs in CI on every change
After this module: Prove a change made the system better rather than merely different.
07Agents and tool use
7 hrs
Agents and tool use
- Planning loops, and where they reliably go wrong
- Tool definitions, MCP and giving a model real capabilities
- Multi-step workflows with checkpoints
- Human-in-the-loop for anything irreversible
- Failure containment, budgets and stopping conditions
After this module: An agent that completes a real task — and stops cleanly when it cannot.
08Production
7 hrs
Production
- Docker, deployment and configuration without secrets in the repo
- Streaming responses and caching what is safe to cache
- Rate limiting, quotas and abuse handling
- Logging, tracing and cost dashboards
- Incident response when the model provider is the outage
After this module: Run the system for real users without a surprise invoice or a silent outage.
Projects
What you will build
Each one exists to prove something specific to someone who is deciding whether to hire you.
Structured-extraction service with validation and retries
You can make model output safe to store.
Hybrid retrieval pipeline with reranking
You understand why RAG systems return the wrong passage.
Capstone: deployed RAG assistant with citations, an eval suite in CI, a cost model and an architecture document
You can build and defend a production AI system. This is the artefact that gets interviews.
Tools you will actually use
How it runs
What a week actually looks like
Live, not recorded
Sessions are taught live so you can interrupt. Recordings are shared afterwards for anyone who misses one.
Small batches
Small enough that everyone gets looked at. If a batch gets large, it splits rather than scales.
Doubt support
A dedicated channel between sessions, and time set aside in each class for questions from the last one.
Cadence
Two to three live sessions a week, evenings and weekends on Indian Standard Time, so it fits around a job.
The certificate
A certificate of completion, issued by NKable, confirming you finished the course and its projects. It certifies completion — not proficiency — and it is not a university qualification. It is not affiliated with, accredited by, or issued on behalf of any university.
What is not included
- Training or fine-tuning foundation models from scratch
- A machine-learning theory course — this is engineering, not research
- Job placement, hiring partners or salary guarantees
- API credits beyond the small budget the coursework needs (roughly the price of a coffee per week, and we cover how to keep it there)
Career support, itemised
So “career guidance” cannot be read as a job promise, here is exactly what it means.
- Capstone architecture review, written up as a portfolio artefact
- Technical interview practice on system-design questions
- CV and GitHub review for AI engineering roles
- An honest read on which roles this course does and does not qualify you for
Full course fee
₹14,999
3 months · ~50.5 live hours · no separate material or exam fees
Enrol in this courseAsk a question firstFees are payable after a short call, not from this page. Refund policy · Enrolment terms
Who teaches this
Nishi Kant Chandra
Based between New Delhi, India and Moscow, Russia
MBA in Business Analytics from O.P. Jindal Global University (O grade), and an M.Sc. in Business Analytics & Big Data Systems from HSE University, Moscow. Google-certified in Data Analytics, with advanced prompt engineering from Vanderbilt. Two and a half years supporting a SaaS platform’s enterprise clients, and the person who built and maintains every tool on this site.
See the full background and the work →AI Engineering — RAG, Agents & Production LLM Systems — questions people ask
How is this different from every other "AI course"?
Module six. Almost no course in this market teaches evaluation — golden sets, retrieval and answer metrics measured separately, groundedness checks, a regression suite in CI. It is the thing that separates "I built a chatbot demo" from "I run an AI system in production", and it is what employers have started screening for.
Do I need to know Python already?
Yes. Functions, classes, packages, the command line and Git. This course does not slow down for the basics, and starting without them means falling behind by week two. If that is you, take Generative AI & Vibe Coding first.
What is the capstone?
A deployed RAG assistant over a document set you choose, with real citations, an evaluation suite running in CI, a cost model and a written architecture document. It is designed to be the artefact you put in front of an interviewer.
Which frameworks do you teach?
Deliberately framework-light: patterns first, then a framework once you understand what it is doing for you. The stack is Python, FastAPI, Postgres with pgvector, an embedding model, one frontier LLM API, Docker and GitHub Actions. Frameworks in this space are replaced faster than a cohort finishes.
Will I need to pay for API credits?
A small amount — roughly the price of a coffee a week if you follow the cost-control practices taught in module two. Keeping that number low is itself part of the syllabus, because it is the same skill that keeps a production system affordable.
Does this cover training or fine-tuning models?
No. This is engineering with foundation models, not training them. Fine-tuning is discussed as an option with its real trade-offs — it is the wrong answer far more often than it is proposed — but you will not train a model in this course.
Will it help me get an AI engineering role?
It gives you a deployed system with evaluation, a written architecture, interview practice on system-design questions, and an honest read on which roles this does and does not qualify you for. There is no placement, hiring partner or salary guarantee attached.
What does the AI Engineering — RAG, Agents & Production LLM Systems course cost?
₹14,999 for the full 3 months, covering roughly 50.5 hours of live teaching plus the projects. There are no separate material or examination fees. Refund terms are on the refund policy page.
How long is the course and how much live teaching is there?
3 months, with about 50.5 hours of live sessions across 8 modules. Sessions are live and online, not recorded lectures you watch alone.
What do I need to know before starting?
Comfortable writing Python (functions, classes, packages). Comfortable with the command line and Git. Able to read API documentation without hand-holding.
Is there a certificate?
Yes — a certificate of completion issued by NKable, confirming you finished the course and its projects. It certifies completion, not proficiency, and it is not a university qualification: it is not affiliated with, accredited by or issued on behalf of any university.
What is not included?
Training or fine-tuning foundation models from scratch; A machine-learning theory course — this is engineering, not research; Job placement, hiring partners or salary guarantees; API credits beyond the small budget the coursework needs (roughly the price of a coffee per week, and we cover how to keep it there). Saying this plainly is deliberate — every course page lists what is included, and almost none say what is not.
Enrol
Join the AI Engineering — RAG, Agents & Production LLM Systems cohort
Message me and tell me what you want to be able to do afterwards. If this course is not the right one for that, I will say so and point you at the one that is.
+91 99991 03353
Opens the chat with your message already written. Usually answered the same day.
[email protected]
Opens your mail app with the subject filled in. Better for anything long.
Nothing is charged from this page. You get a reply with the next batch date, the schedule and the fee, and payment happens separately after a conversation.