You'll use both proprietary APIs and open-source models. No single vendor lock-in — you'll know how to choose the right model for any task and budget.
GPT-4o
The benchmark model for reasoning, coding, and multimodal tasks. Used for API pattern learning and production comparison.
✓ Weeks 1, 3, 6Mistral 7B / Mixtral
Best open-source alternative. Fine-tunable locally, deployable on your own infra. Used for RAG and fine-tuning projects.
✓ Weeks 2, 4, 5Llama 3
Meta's flagship open model. Used for instruction fine-tuning with LoRA and QLoRA on domain-specific tasks.
✓ Week 4Gemini 1.5 Flash
Long-context champion. 1M token context window. Used for document Q&A and summarisation over massive corpora.
✓ Week 3Claude 3 Haiku
Fast, cost-efficient, with strong instruction following. Used for high-throughput classification and agent tasks.
✓ Week 5Open Model Hub
phi-3, Gemma-2, Qwen2, Falcon — the full ecosystem. You'll know how to evaluate, select, and deploy any open model.
✓ ThroughoutRAG (Retrieval-Augmented Generation) is the most deployed GenAI architecture in production. You build it from scratch and understand every component.
Document Loading
PDFs, URLs, Notion, APIs — LangChain loaders for any source.
Text Splitting
Recursive, semantic, and sentence-aware chunking strategies.
Vector Embeddings
OpenAI / SBERT embeddings stored in ChromaDB or Pinecone.
Similarity Search
Top-k retrieval, MMR reranking, hybrid sparse+dense search.
LLM Response
Augmented prompt → grounded, accurate, cited answer.
You'll learn to construct, evaluate, and improve prompts systematically. Here are three techniques you'll master — shown with real examples:
GenAI Engineer
The hottest job title in tech right now. Building LLM-powered products, RAG systems, and agents at AI-first companies.
₹15–35 LPA fresher rangeAI Agent Developer
Autonomous agent systems are being deployed everywhere. LangGraph and tool-calling skills are the entry ticket.
₹18–40 LPA rangeLLM Application Engineer
Building on top of foundation models at product companies — chatbots, copilots, search, summarisation tools.
₹14–30 LPA rangeAI Researcher / Fine-tuner
LoRA/QLoRA fine-tuning skills open roles at research labs and companies adapting models for specific domains.
₹20–50 LPA range// This course is for
- How LLMs work: decoder-only transformers, autoregressive generation
- Tokenisation: BPE, tiktoken, context window and token cost
- Temperature, top-p, top-k — controlling output diversity
- OpenAI API: chat completions, system/user/assistant roles, streaming
- Zero-shot, few-shot, chain-of-thought prompting — with eval benchmarks
- ReAct prompting and self-consistency
- Structured outputs: JSON mode, function calling, Pydantic models
- Prompt injection attacks and defence patterns
- Text embeddings: what they are, why they work, dimensions
- OpenAI text-embedding-3, SBERT — comparison and selection
- ChromaDB: store, query, filter, update vectors
- Pinecone: managed vector DB for production scale
- Cosine similarity, dot product — the maths of retrieval
- Semantic search engine: index your own document corpus
- Hybrid search: BM25 + dense retrieval combined
- Reranking: Cohere Rerank, cross-encoder models
- LangChain: chains, document loaders, text splitters, retrievers
- LlamaIndex: query engines, node parsers, indices
- Basic RAG: load → chunk → embed → retrieve → generate
- Advanced RAG: HyDE, multi-query retrieval, contextual compression
- Conversational RAG: memory, chat history, follow-up questions
- Long-context RAG with Gemini 1.5 Flash (1M token window)
- RAGAS evaluation: faithfulness, answer relevancy, context precision
- Project: Intelligent Q&A system over your own PDF library
- When to RAG vs when to fine-tune — decision framework
- Full fine-tuning vs PEFT (parameter-efficient) methods
- LoRA: low-rank adaptation — theory, rank, alpha, target modules
- QLoRA: 4-bit quantisation + LoRA for consumer hardware
- Instruction dataset format: system/user/assistant JSONL
- Fine-tune Llama 3 on domain data with HuggingFace PEFT + TRL
- Evaluation: perplexity, task-specific metrics, human eval
- Merge adapter weights and push to HuggingFace Hub
- Agent architecture: LLM + tools + memory + planning
- Tool calling: function definitions, JSON schema, OpenAI tools API
- LangChain agents: ReAct, OpenAI Functions agent
- LangGraph: stateful multi-step agent workflows, conditional edges
- Memory types: conversation buffer, summary, vector store memory
- Build: a research agent that searches, reads, and summarises web content
- Multi-agent systems: supervisor-worker patterns with LangGraph
- Agent safety: output validation, loop detection, cost guardrails
- Multimodal LLMs: GPT-4o vision, LLaVA, Gemini Pro Vision
- Image + text pipelines: describe, extract, and reason over images
- Streaming responses: Server-Sent Events, Gradio stream, FastAPI
- Cost optimisation: caching, model routing, prompt compression
- Latency and throughput: batching, async calls, vLLM basics
- LLM guardrails: Nemo Guardrails, input/output filtering
- Capstone: production GenAI application of your choice
- Capstone demo to batch + mentor review + Gold badge issuance
Prompt Engineering Benchmark
Systematically compare 5 prompting techniques (zero-shot to CoT to ReAct) on 3 tasks. Build an evaluation harness that measures accuracy, consistency, and cost per task.
Document Intelligence RAG App
Ingest a corpus of 50+ PDFs (research papers, policies, contracts). Build a conversational Q&A interface with citation tracking, source attribution, and RAGAS-evaluated faithfulness.
Domain-Specific Fine-Tuned Model
Curate an instruction dataset, fine-tune Llama 3 with QLoRA on a specific domain (medical, legal, or finance), evaluate against the base model, and publish to HuggingFace Hub.
Autonomous Research Agent
A LangGraph-powered agent that takes a research question, autonomously searches the web, reads relevant pages, synthesises findings, and produces a structured report with citations.
Newton JEE Gold Badge
NLP & GenAI Specialist — Generative AI & LLMs
The Two-Gold-Badge Combination
The GenAI Gold badge, combined with the NLP Gold badge, creates a uniquely powerful LinkedIn credential cluster: NLP & GenAI Specialist. This combination is the most recruiter-visible signal for LLM engineering and GenAI product roles in the current market.
Every session felt like being inside a real GenAI engineering team. Priya doesn't teach you to call an API — she teaches you to think about why you're calling it, when to switch models, and how to debug when it gives nonsense. I shipped my first RAG app in week 3. It's in production now.
The fine-tuning week was a revelation. I had thought fine-tuning required a full GPU cluster. QLoRA on Colab was a complete paradigm shift. I fine-tuned Llama 3 on medical Q&A, published it to HuggingFace, and it's now my most-starred repo with 200+ downloads.
The agent week was the most mind-expanding week of any course I've taken. Building a research agent that autonomously reads, plans, and synthesises — and seeing it actually work — felt like the future arriving early. Got 3 interview calls the week after posting the demo on LinkedIn.
5.0 rating from me. The capstone process — where you define the problem, build it, deploy it, and demo it live to the batch — is unlike any other learning experience. Priya's feedback during the demo was surgical. My capstone is now my portfolio's centrepiece and it led directly to my current role.