AI Engineering — Building Apps with Large Language Models

Build real applications on top of large language models: how LLMs work, calling the Claude API from Python, prompt engineering, structured JSON output, conversations and streaming, embeddings and semantic search, retrieval-augmented generation (RAG), tool use and agents, evaluation, cost and safety — with Indian case studies and a capstone project. Logic is run for real; API calls are checked against a local stand-in for the API, which you can use to practise for free.

1. 01How Large Language Models Work — Tokens, Next-Token Prediction, Context Windows and LimitsWhat an LLM really does (predict the next token), tokens and context windows, training vs using a model, why models make things up, what they're good and bad at, and what an "AI engineer" builds around them — with a tiny runnable next-word predictor.
2. 02Your First LLM API Call — Keys, the Python SDK, Responses, Tokens and CostCall an LLM from Python with the official Claude SDK: keeping API keys safe, the anatomy of a request and response, stop reasons, token usage, estimating cost in rupees, handling errors and retries — and how to practise for free with a local stand-in API.
3. 03Prompt Engineering That Works — Clear Instructions, Context, Examples and Output FormatsWrite prompts like an engineer: role and goal, giving context, specific instructions, examples (few-shot), tagging data, output formats, prompt templates in code, and testing prompts — with before/after examples from Indian use cases.
4. 04Structured Outputs — Getting Reliable JSON from an LLM with Schemas and PydanticTurn messy text into data your program can use: JSON schemas, the API's structured output mode, Pydantic models, validating results, and a resume-parsing case study — plus why "please reply in JSON" in the prompt is not enough.
5. 05Chatbots — Conversation History, Memory Limits and Streaming ResponsesBuild a multi-turn chatbot: keeping conversation history, why it grows and costs more each turn, trimming and summarising old turns, and streaming the answer word by word so users don't stare at a spinner.
6. 06Embeddings and Semantic Search — Finding Text by Meaning, Vector DatabasesHow computers find text by meaning: keyword search vs semantic search, vectors and cosine similarity, a runnable search over a college handbook, how embedding models and vector databases (pgvector, Chroma) scale it up, and hybrid search.
7. 07RAG — Retrieval-Augmented Generation with Chunking, Citations and EvaluationBuild a question-answering bot over your own documents: chunking, retrieving the best passages, a grounded prompt with numbered sources, citations, saying "not in the documents", and measuring retrieval quality — through a college handbook case study.
8. 08Tool Use and AI Agents — Letting the Model Call Your Functions SafelyGive an LLM tools: define them with JSON schemas, run the tool-use loop, return results and errors, and decide when you need a full agent versus a simple workflow — with permission checks, step limits and human approval for risky actions.
9. 09Evaluating LLM Apps — Test Sets, Scoring, LLM-as-Judge and Regression TestingStop judging AI features by "looks good": build a test set from real questions, score with rules and reference answers, use an LLM as a judge carefully, compare versions before shipping, and keep evaluating in production.
10. 10LLM Apps in Production — Cost, Latency, Prompt Caching, Rate Limits and MonitoringRun AI features reliably and affordably: where the cost comes from, prompt caching, choosing models per task, batching, rate limiting your users, timeouts and fallbacks, handling refusals, and what to log — with a cost model you can run.
11. 11Safe and Responsible AI Apps — Prompt Injection, Privacy, Bias and Human OversightThe risks every AI engineer must design for: prompt injection, leaking private data, over-trusting outputs, bias and misuse — with practical defences, a runnable personal-data redactor, India's DPDP Act basics, and when a human must stay in the loop.
12. 12Capstone — Build a Campus Help-Desk Assistant, Plus AI Engineering Interview QuestionsPlan and build a complete AI project — a campus help-desk assistant with RAG, tools, guardrails and evaluation — then revise the AI engineering interview questions asked for LLM, GenAI and AI developer roles.