Home › Courses › Generative AI Course in Pune
3-month practical Generative AI programGenerative AI Course in Pune with LLMs, RAG & AI Agents
Learn how tools like ChatGPT really work, then build your own. In 12 weeks you go from neural network basics to fine-tuning models, building RAG apps and AI agents, and deploying them.
- 36 live lectures
- 75% hands-on coding
- Agentic AI included
- ISO certificate
- Classroom or online
- Placement assistance
Your 12 weeks at a glance
36 lectures · 2 hours each · 3 lectures a week · 72+ hours
Sit in on a free demo class
See a live AI lab, meet the trainer and get the next batch dates. Our team calls you back.
About the course
Generative AI Classes in Pune at Our Chinchwad Centre
Teknowell's Generative AI (Gen AI) course in Pune is a 3-month, code-first program for people who want to build with AI, not just use it. You learn how large language models (LLMs) are built and trained, then use that knowledge to create real applications: assistants that answer from your own documents, fine-tuned models for specific tasks, and AI agents that use tools to get work done.
The course has 36 live lectures of 2 hours each. From Lecture 4 onwards, about 75% of class time is spent coding in Google Colab or VS Code, and every module ends with something you have built and can show an employer.
Generative AI training in Pune, online or in the classroom
Classes run at our centre in Kunal Plaza, near Chinchwad Station, and live online for students anywhere in Pune. With 8 batch slots between 9:30 AM and 8:30 PM, college students and working professionals can both join.
Wondering how AI will change software jobs? Read our guide, Will AI Replace Software Developers in India?
What is Generative AI?
Generative AI is a type of artificial intelligence that creates new content, such as text, code or images, by learning patterns from large amounts of data. Tools like ChatGPT are built on large language models (LLMs), which predict the next word to write fluent, useful answers.
Generative AI vs Agentic AI
Generative AI produces content when you ask for it. Agentic AI uses that ability to act: an AI agent breaks a goal into steps, calls tools such as search or code, checks the results and keeps going until the task is done. This course teaches both.
What you learn
You Learn the Whole Generative AI Stack
Most courses teach only the top layer. You work through every layer, from the model itself to the apps built on it.
Layer 5 · Apps
Agentic AI & Prompt Engineering
AI agents that plan and use tools, multi-agent workflows, memory, AI coding agents and prompts for real code bases.
Layer 4 · Retrieval
RAG Training
Connect AI to your own documents with embeddings, vector databases, ColBERT, GraphRAG and answer evaluation.
Layer 3 · Adaptation
Fine-Tuning LLMs & Alignment
LoRA and QLoRA on Qwen, Llama and Phi, vision models that read documents, and RLHF, DPO and GRPO for reasoning.
Layer 2 · Models
LLMs & Transformers
Tokenization, embeddings, self-attention and training. You build a small transformer from scratch.
Layer 1 · Infrastructure
Inference & Deployment (LLMOps)
Quantization, fast serving with vLLM and Ollama, Docker, FastAPI, monitoring and security.
Is this course for you?
A Quick Fit Check
Great fit if you are...
- An engineering student in CS, IT, AI or Data Science who wants a GenAI portfolio before graduating
- A graduate or fresher with basic Python who wants an AI engineering role
- A working developer or analyst adding AI skills (evening batches until 8:30 PM)
- Returning after an education gap and ready to put in 3 lectures a week
Not the right fit if you...
- Only want to use ChatGPT for office work. Our short Generative AI upskilling course suits you better.
- Want theory only, with no coding. This course is about 75% hands-on.
- Can't set aside time for weekly labs. Projects are how you are graded.
Generative AI syllabus
9 Modules, 36 Lectures, 9 Things You Ship
Module 1 of 9
Generative AI Foundations & Core Architectures
You shipCustom BPE tokenizer and a PyTorch attention block
Lecture 01 · 2 hrs
Introduction to Generative AI & Market Landscape
- Evolution from Rules-based NLP to Statistical LM to Generative AI.
- Overview of LLM Stack: Compute, Base Models, Fine-Tuning, RAG, and Agents.
- Understanding Open-Weight vs. Closed API Ecosystem (DeepSeek, Qwen, Llama vs OpenAI, Anthropic).
- Hardware 101: GPU VRAM sizing, FLOPS, memory bandwidth, and quantization necessity.
LabSetup Python 3.11+, PyTorch CUDA environment on Google Colab / Local GPU. Run benchmark tests on VRAM memory throughput.
Lecture 02 · 2 hrs
Neural Network Prerequisite Refresher & Tensor Ops
- PyTorch Fundamentals: Tensors, Autograd, Backpropagation intuition.
- Linear layers, activation functions (ReLU, GELU, SwiGLU), and LayerNorm vs RMSNorm.
- Matrix multiplication mechanics: Shape matching, batch processing, matrix dimensions.
LabBuild a simple multi-layer perceptron (MLP) from scratch in PyTorch to classify synthesized embeddings.
Lecture 03 · 2 hrs
Tokenization Deep Dive & Modern Embeddings (ColBERT Concept)
- How text becomes numbers: Character, Word, Subword Tokenization (BPE, WordPiece, Unigram).
- Tokenizers in practice: Tiktoken, SentencePiece, HuggingFace Tokenizers.
- Dense Embeddings vs. Sparse Embeddings.
- Introduction to ColBERT (Contextualized Late Interaction over BERT): token-level dense representation vs single-vector pooling.
LabBuild a basic Byte-Pair Encoding (BPE) tokenizer from scratch in Python. Compare output of OpenAI tiktoken vs Llama tokenizer. Implement a toy ColBERT late-interaction scoring routine.
Lecture 04 · 2 hrs
The Transformer Architecture: Self-Attention Mechanics
- Query, Key, Value (Q, K, V) vector intuitive explanation.
- Scaled Dot-Product Attention: Math, scaling factor sqrt(d_k), and Softmax.
- Multi-Head Attention (MHA), Grouped Query Attention (GQA), and Multi-Query Attention (MQA).
- Positional Encodings: Absolute, Sinusoidal, and Rotary Position Embeddings (RoPE).
LabWrite Scaled Dot-Product Attention and Multi-Head Attention modules from scratch using raw PyTorch tensor operations.
Module 2 of 9
Building & Training an LLM from Scratch
You shipA working NanoGPT-style transformer, trained on Colab
Lecture 05 · 2 hrs
Building the Decoder Block & Causal Masking
- Decoder-only vs Encoder-Decoder vs Encoder-only architectures.
- Causal Masking: Why upper-triangular matrix is filled with -inf.
- Residual connections and pre-LN vs post-LN layer normalization.
- Feed-Forward Network (FFN) with SwiGLU activation.
LabAssemble the Transformer Decoder Block in PyTorch by combining Masked MHA, RMSNorm, and SwiGLU FFN.
Lecture 06 · 2 hrs
Assembling NanoGPT & Pre-training Pipeline
- Complete model architecture definition: Token Embedding, Positional Embedding, N-Decoder Blocks, Final LM Head.
- Loss Function: Cross-Entropy Loss over target tokens, Perplexity metric.
- Data Pipeline: Tokenizing datasets (TinyStories / FineWeb-Edu subset) into memory-mapped arrays.
LabComplete full PyTorch code for a 10M parameter NanoGPT model. Implement data loader for autoregressive training.
Lecture 07 · 2 hrs
Training Dynamics, Optimizers & Precision
- AdamW Optimizer: Learning rate schedules, warmup, and cosine decay.
- Mixed Precision Training: FP32 vs FP16 vs BF16, GradScaler usage.
- Gradient Clipping, Weight Decay, and preventing exploding/vanishing gradients.
LabTrain the 10M NanoGPT model on TinyStories dataset for 1000 steps in Colab. Track training loss and plot learning rate curve.
Lecture 08 · 2 hrs
Autoregressive Text Generation & Decoding Strategies
- Generation loop: KV-Cache intuition (saving computed Keys and Values).
- Decoding strategies: Greedy search, Beam search.
- Stochastic sampling: Temperature scaling, Top-K sampling, Top-P (Nucleus) sampling, and Repetition Penalty.
LabWrite the inference generation script from scratch. Implement KV-Cache in Python. Test text generation with varying Temperature and Top-P values.
Module 3 of 9
Advanced RAG, ColBERT & TurboVec
You shipA production RAG pipeline with multi-vector retrieval
Lecture 09 · 2 hrs
RAG Architecture & Document Processing Pipeline
- Naive RAG vs Advanced RAG vs Modular RAG.
- Document Ingestion: Chunking strategies (Fixed size, Recursive character, Semantic chunking).
- Embedding Generation: Bi-encoders (OpenAI, BGE, E5).
- Vector Databases: Index types (HNSW, IVFFlat) and similarity metrics (Cosine, L2, Inner Product).
LabBuild a basic RAG pipeline using LangChain / LlamaIndex and ChromaDB to query PDF documents.
Lecture 10 · 2 hrs
Multi-Vector Retrieval & ColBERT Implementation
- Limitations of single-vector dense embeddings for long context/complex queries.
- ColBERT Late Interaction: MaxSim operator mechanics.
- RAGatouille library: Training/Using ColBERTv2 models easily.
- Hybrid Search: Combining BM25 keyword search with Dense & ColBERT embeddings via Reciprocal Rank Fusion (RRF).
LabImplement RAGatouille (ColBERTv2) in Python for document retrieval. Build a hybrid search pipeline with RRF reranking.
Lecture 11 · 2 hrs
Ultra-Fast Vector Search: TurboVec & Vector Indexing
- High-performance vector search: Demystifying SIMD, AVX-512, and GPU-accelerated vector indexing.
- TurboVec Architecture: High-throughput, low-latency vector indexing techniques for large-scale embeddings.
- Comparison of vector backends: TurboVec vs Faiss vs Milvus vs Qdrant in latency and recall.
LabBenchmark vector retrieval speed across 100,000 vectors comparing standard Python cosine similarity vs Faiss vs TurboVec-enabled vector index.
Lecture 12 · 2 hrs
Contextual Retrieval, GraphRAG & Evaluation
- Advanced RAG Techniques: Contextual Compression, Parent Document Retriever, HyDE.
- GraphRAG: Knowledge Graphs combined with Vector Search for multi-hop reasoning.
- RAG Evaluation: Ragas framework (Faithfulness, Answer Relevance, Context Recall, Context Precision).
LabBuild a production RAG system with HyDE and evaluate performance using the Ragas framework on a synthetic QA dataset.
Module 4 of 9
Fine-Tuning, SLMs & Vision-Language OCR
You shipA fine-tuned small language model that returns structured JSON
Lecture 13 · 2 hrs
Fine-Tuning Fundamentals & Parameter-Efficient Fine-Tuning (PEFT)
- Full Fine-Tuning vs PEFT: Why full tuning is computationally expensive.
- LoRA (Low-Rank Adaptation): Math behind rank matrices (A and B), alpha parameter, target modules.
- QLoRA: 4-bit NormalFloat (NF4) quantization, Double Quantization, Paged Optimizers.
LabConfigure LoRA parameters in PyTorch. Implement a micro-LoRA layer from scratch on a linear matrix.
Lecture 14 · 2 hrs
Hands-on Fine-Tuning Open SLMs (Qwen / Llama / Phi)
- Small Language Models (SLMs): Why 1B-3B models rule edge computing.
- Data Preparation: Instruction formatting (ChatML, Alpaca format, System prompts).
- SFT using Hugging Face TRL (SFTTrainer) and Unsloth for 2x-5x speedup.
LabUse Unsloth / Hugging Face SFTTrainer on Google Colab GPU to fine-tune Qwen 2.5 / Llama 3 3B on a custom instruction dataset.
Lecture 15 · 2 hrs
Vision-Language Models (VLMs) & Multimodal OCR
- Multimodal Architectures: Vision Encoder (ViT / CLIP) + Projection Layer + Language Model.
- Multimodal Projector: Linear layers vs Cross-Attention vs MLP connectors.
- Document Intelligence: Using VLMs (Qwen-VL, GLM-Vision) for structured data extraction.
LabLoad Qwen-VL / Phi-3-Vision in Colab. Pass invoice/cheque images and extract structured JSON data matching a Pydantic schema.
Lecture 16 · 2 hrs
Fine-Tuning VLMs for Structured JSON Extraction
- Preparing Multimodal Datasets: Image-Text pairs with structured target JSON schemas.
- Enforcing valid JSON output: Outlines, Instructor, and JSON-Mode generation parameters.
- Handling noisy scanned OCR images and low-resolution inputs.
LabFine-tune a small VLM (Qwen-2.5-VL-3B or Florence-2) to extract custom fields from unstructured scanned documents into clean JSON.
Module 5 of 9
Inference Optimization: MTP, dFlash & Quantization
You shipA fast, quantized model served with vLLM
Lecture 17 · 2 hrs
Model Quantization Techniques (GGUF, AWQ, GPTQ, EXL2)
- Quantization Fundamentals: PTQ vs QAT.
- Symmetric vs Asymmetric Quantization, INT8, INT4, FP4.
- Formats: GGUF (llama.cpp format), AWQ, GPTQ, EXL2.
LabQuantize an FP16 open-weight model to GGUF (Q4_K_M) using llama.cpp scripts. Run benchmark tests comparing latency and memory footprints.
Lecture 18 · 2 hrs
Speculative Decoding & Multi-Token Prediction (MTP)
- Inference Bottlenecks: Memory-bandwidth bound vs Compute-bound generation.
- Speculative Decoding: Using a small Draft Model to verify tokens with a Target Model.
- MTP: DeepSeek V3 innovation of predicting N future tokens simultaneously.
LabImplement a toy Speculative Decoding loop in PyTorch with a 135M draft model and 1B target model. Measure latency speedup factor.
Lecture 19 · 2 hrs
dFlash Attention & Modern Attention Acceleration
- Attention Complexity Revisited: O(N^2) memory bottleneck in long contexts.
- FlashAttention-1, 2, 3: Tiling, IO-awareness, SRAM vs HBM memory management.
- dFlash Attention: Skipping non-essential attention blocks for ultra-fast long-context processing.
LabProfile standard PyTorch attention vs FlashAttention vs dynamic sparse attention. Measure VRAM and speed across context lengths (1k to 32k).
Lecture 20 · 2 hrs
Production Inference Servers (vLLM, Ollama, TGI)
- PagedAttention & vLLM: Virtual memory allocation for KV-Cache.
- Continuous Batching vs Static Batching.
- Deploying local REST API endpoints with vLLM, Ollama, and TensorRT-LLM.
LabSpin up a local vLLM / Ollama server, configure API endpoints, and run load testing using Python asyncio and httpx.
Module 6 of 9
Autonomous AI Agents & Tool Calling
You shipA multi-agent system with tool calling and memory
Lecture 21 · 2 hrs
Agentic Architectures & ReAct Framework
- What makes an Agent? Model + System Prompt + Tools + Memory + Loop.
- ReAct Pattern: Reason -> Act -> Observe -> Repeat loop.
- Planning & Reflection techniques (Tree of Thoughts, Self-Refine).
LabBuild a complete ReAct Agent loop from scratch in pure Python without using any external framework.
Lecture 22 · 2 hrs
Function Calling & JSON Tool Schema Integration
- Native Function Calling: How LLMs output structured tool invocation calls.
- OpenAI / Anthropic / Ollama Tool Calling Spec: JSON Schemas, parameter validation.
- Handling tool call execution results and feeding responses back to the model context.
LabDefine custom Python functions (Weather API, Database Query, Calculator). Bind them via JSON schemas to an open LLM and execute automated tool loops.
Lecture 23 · 2 hrs
Multi-Agent Frameworks & Collaborative Workflows
- Multi-Agent Patterns: Hierarchical, Sequential, Network, and Supervisor setups.
- Frameworks overview: LangGraph, CrewAI, AutoGen.
- State Management, Persistence, and Human-in-the-Loop approval workflows.
LabBuild a 3-agent research system in LangGraph / CrewAI (Web Researcher, Technical Summarizer, Code Reviewer).
Lecture 24 · 2 hrs
Agent Memory, Vector Persistence & Sandboxing
- Short-term memory (Context Window) vs Long-term memory (Vector Store / Key-Value).
- Agent state persistence across sessions.
- Execution Safety: Sandboxing code execution environment (Docker / E2B code interpreter).
LabIntegrate E2B / Docker code execution sandbox into your agent so it can safely execute arbitrary Python code generated during problem solving.
Module 7 of 9
Developer Workflows & CLI Coding Harnesses
You shipA feature built end to end with AI coding tools
Lecture 25 · 2 hrs
Introduction to AI Coding Harnesses & Agentic CLI Tools
- The Shift in Software Engineering: From writing code to orchestrating AI coding agents.
- Overview of CLI Harnesses: Claude Code, OpenCode, Gemini CLI, Cursor, Aider.
- How coding harnesses index codebases: ASTs, Tree-sitter, Git diffs, repo-maps.
LabInstall and set up Claude Code / OpenCode / Aider CLI environments. Index a sample 5,000-line GitHub repository.
Lecture 26 · 2 hrs
Prompt Engineering & System Directives for Code Bases
- Writing effective .clauderc, AGENTS.md, and custom system instructions for repository context.
- Context window management: Token budgets, file pruning, and repo mapping.
- Architectural guidance to prevent model hallucination in large codebases.
LabCreate a full system harness definition (AGENTS.md) for an existing Flask/FastAPI project and instruct the CLI harness to implement unit test suites autonomously.
Lecture 27 · 2 hrs
Automated Code Generation, Refactoring & Bug Fixing
- Automated PR Generation, multi-file refactoring, and test-driven agent loops.
- Tool integration with Git: Automated branch creation, committing, and conflict resolution via agent harness.
- Benchmarking Coding Agents: HumanEval, SWE-bench evaluation methodology.
LabUse a CLI coding harness to automatically debug a broken open-source repository, pass failing test suites, and draft a pull request.
Lecture 28 · 2 hrs
Custom Harness Construction & Model Context Protocol (MCP)
- Model Context Protocol (MCP): Open standard for connecting AI models to local data sources and tools.
- Building custom MCP servers in Python/TypeScript.
- Integrating local IDEs and terminal harnesses with external APIs and databases via MCP.
LabBuild a custom MCP server in Python that connects a CLI harness directly to a local PostgreSQL database for schema analysis.
Module 8 of 9
Reinforcement Learning (RLHF, DPO & GRPO)
You shipAn LLM tuned with DPO or GRPO for better reasoning
Lecture 29 · 2 hrs
Post-Training Pipeline: SFT vs Alignment RL
- Why Base Models aren't Assistant Models: Pre-training vs Post-training.
- Alignment Problem: Helpful, Honest, and Harmless (HHH).
- RLHF Workflow: SFT -> Reward Model Training -> PPO Reinforcement Learning.
LabAnalyze response preference datasets (Anthropic HH-RLHF). Train a simple Reward Model in PyTorch using binary cross-entropy loss over preferred vs rejected pairs.
Lecture 30 · 2 hrs
Direct Preference Optimization (DPO) & Direct Alignment
- Mathematical simplification of RLHF: Eliminating the explicit Reward Model.
- DPO Loss Function: Implicit reward derivation from reference model and policy model.
- Variants: KTO (Kahneman-Tversky Optimization), ORPO, IPO.
LabFine-tune a 1B-3B SLM using DPO with Hugging Face TRL DPOTrainer to align conversational tone and eliminate refusal hallucinations.
Lecture 31 · 2 hrs
Group Relative Policy Optimization (GRPO) & Reasoning LLMs
- The DeepSeek-R1 Revolution: Rule-based reinforcement learning without reliance on traditional reward models.
- GRPO Algorithm: Group sampling of trajectories, relative advantage normalization.
- Reward Functions: Enforcing chain-of-thought <think> tags, mathematical accuracy, formatting compliance.
LabImplement a toy GRPO reward function in Python to reward model outputs for correctly solving math logic puzzles step-by-step.
Lecture 32 · 2 hrs
Practical RL Alignment Workshop
- Setting up an end-to-end alignment pipeline on Colab.
- Evaluating aligned models: Win-rate calculation using LLM-as-a-Judge (GPT-4 / Claude as evaluator).
- Preventing reward hacking and mode collapse.
LabExecute a full DPO/GRPO alignment notebook. Evaluate pre-aligned vs post-aligned model outputs using an automated LLM-as-a-Judge script.
Module 9 of 9
Capstone Project & Industry Deployment
You shipA production AI app deployed with Docker and an API
Lecture 33 · 2 hrs
Production System Architecture & API Design
- Designing enterprise-grade GenAI systems: FastAPI wrapper, queue systems (Celery/Redis), rate limiting.
- Streaming responses: Server-Sent Events (SSE) and WebSockets for real-time token streaming.
- Cost calculation & latency optimization strategies.
LabBuild a FastAPI backend service that streams LLM tokens via SSE and exposes endpoints for RAG and tool execution.
Lecture 34 · 2 hrs
Containerization & Cloud Infrastructure (Docker & GPU)
- Dockerizing AI applications: CUDA base images, dependency management, weight mounting.
- Deploying to Cloud GPU providers (RunPod, Modal, AWS EC2 g5, GCP Vertex).
- Security: Environment variables, API key management, static public IP whitelisting.
LabWrite a Dockerfile for your FastAPI + vLLM / RAG app. Package container and run it locally, verifying GPU pass-through (--gpus all).
Lecture 35 · 2 hrs
Observability, Evaluation & Security (LLMOps)
- LLMOps stack: Monitoring latency, token usage, drift, and hallucination metrics.
- Observability tools: LangSmith, Phoenix (Arize), OpenInference.
- AI Security: Prompt Injection mitigation, PII masking, guardrails (NeMo Guardrails, Llama Guard).
LabIntegrate LangSmith / Phoenix observability into your application. Add a Llama Guard safety filter layer to block prompt injection attacks.
Lecture 36 · 2 hrs
Capstone Project Showcase & Industry Readiness
- Final Capstone Presentation: Demonstration of end-to-end industry-ready GenAI product.
- Resume Building for GenAI Roles: Key project bullet points, GitHub portfolio curation.
- Industry Technical Interview Prep: Architecture design whiteboarding, coding round expectations.
LabLive code demonstration of capstone projects to industry mentors / instructors. Receive feedback and portfolio reviews.
Your portfolio
9 Projects on Your GitHub Before You Graduate
One working project per module. Walk any interviewer through them.
your-github / tokenizer-from-scratchModule 1
Custom BPE tokenizer and a PyTorch attention block.
PyTorchBPEAttention
your-github / nanogpt-colabModule 2
A working NanoGPT-style transformer, trained on Colab.
PyTorchTransformerPre-training
your-github / rag-engineModule 3
A production RAG pipeline with multi-vector retrieval.
PythonQdrantColBERTRagas
your-github / doc-to-json-slmModule 4
A fine-tuned small language model that returns structured JSON.
PythonQwenLoRAVLM
your-github / fast-llm-servingModule 5
A fast, quantized model served with vLLM.
PythonvLLMGGUFQuantization
your-github / multi-agent-assistantModule 6
A multi-agent system with tool calling and memory.
PythonLangGraphToolsMemory
your-github / ai-coding-workflowModule 7
A feature built end to end with AI coding tools.
TypeScriptClaude CodeMCP
your-github / reasoning-dpo-grpoModule 8
An LLM tuned with DPO or GRPO for better reasoning.
PythonTRLDPOGRPO
your-github / genai-capstoneModule 9
A production AI app deployed with Docker and an API.
PythonFastAPIDockerAgents
How you are assessed
Graded on a Portfolio, Not Exams
- 30%Weekly lab notebooks
- 20%Mid-term RAG engine
- 20%Fine-tuned document extractor
- 30%Final capstone product
Tools & technologies
The Stack Production AI Teams Use
Models & training
Hugging Face
UnslothRetrieval (RAG)
Faiss
Qdrant
TurboVec
RagasAgents
LangGraph
CrewAI
LangSmithServing & deployment
vLLM
Ollamallama.cppllama.cpp
FastAPIDocker
E2BAI coding tools
Claude Code
Gemini CLI
OpenCode
AiderFees & batches
Generative AI Course Fees and Batch Timings in Pune
Short answer: The Generative AI course fee at Teknowell depends on whether you choose classroom or online training and on current batch offers. Call or WhatsApp +91 91460 11401 for the exact fee. Batches run in 8 slots between 9:30 AM and 8:30 PM.
What the fee includes
- 36 live lectures
- Weekly hands-on labs
- 9 module projects
- Capstone product
- Mock interviews
- Placement assistance
- ISO-certified certificate
- Internship certificate
- Duration
- 3 months (12 weeks)
- Next batch
- New batches start regularly. Ask us for the next date.
- Fees
- Shared on call or WhatsApp
Careers
Generative AI Jobs in Pune After This Course
Short answer: After a Generative AI course you can apply for roles such as Generative AI Engineer, LLM Application Developer, AI Agent Developer, RAG Engineer and Prompt Engineer. In Pune, these roles are growing across IT services companies, product firms and startups in Hinjawadi, Kharadi, Magarpatta and Baner.
Builds and ships LLM features into products.
Creates chat assistants and AI tools on top of models.
Builds agents that plan, call tools and complete tasks.
Connects AI to company documents and data.
Designs and tests prompts and system instructions.
Deploys, monitors and optimises models in production.
Why choose us
Why Students Choose Teknowell for Generative AI
Up-to-date syllabus
RAG, fine-tuning, AI agents, MCP and modern inference, as teams use them in 2026.
Code-first classes
About 75% of class time is spent coding, with a lab in every lecture.
Placement assistance
Resume help, mock interviews, hiring connections and an internship alongside training.
Student reviews
What Our Students Say on Google
Placements
Our Placed Students
Teknowell graduates now working at companies across Pune and beyond.
Akash Sawant
Software Developer
Hired byInfrabeat Technologies Pvt Ltd
Manjiri Chaudhari
Software Developer
Hired byInfrabeat Technologies Pvt Ltd
Deependra Rawat
Associate Data Analyst
Hired byTeknoverse IT Solutions
Kalpesh Sonawane
Project Engineer
Hired byWipro
Preeti Shinde
Software Developer
Hired byCloud Booking Software
Pranav Kate
Software Developer
Hired byITCubeHiring partners
Where Our Students Get Hired
Top firms across startups, IT services and product companies.
Career ready
Ready to Start Your Career in Generative AI?
Sit in on a free demo class, see a live AI lab and get the next batch dates. No commitment.
Why Choose Teknowell for Generative AI Courses in Pune
Generative AI has moved from research labs into everyday software in just a few years, and companies in Pune now hire for it across IT services, product firms and startups. Most students looking for a Generative AI course in Pune want the same thing: to understand how tools like ChatGPT actually work, and to build real AI applications that they can show in an interview. The Generative AI course at Teknowell EduTech is designed around exactly that. Over 3 months and 36 live lectures, you go from the basics of neural networks to building, fine-tuning and deploying your own AI applications, with about 75% of class time spent writing code.
What a good Generative AI course in Pune should teach
Plenty of courses stop at prompt tricks and API calls. That is useful, but it is not enough for an engineering job. A strong Gen AI course should explain how large language models work, show you how to connect them to your own data, teach you to customise them, and prepare you to deploy them safely. Our syllabus follows that path. You start with transformers and tokenization, build a small language model from scratch, and then move on to RAG, fine-tuning, AI agents and production deployment.
LLM training that starts with how models really work
In the first two modules you build the core pieces yourself: a tokenizer, a self-attention block and a small NanoGPT-style transformer trained on Google Colab. You learn why models need so much GPU memory, how training is tuned, and how text generation settings such as temperature and top-p change the output. Understanding these basics makes every later topic easier, and it gives you clear answers when an interviewer asks how an LLM works.
RAG training: connect AI to your own documents
Retrieval-Augmented Generation, or RAG, is one of the most requested Generative AI skills in companies today, because it lets an AI answer questions from a company's own documents. In Module 3 you process PDFs, create embeddings, store them in vector databases such as Qdrant and Faiss, improve retrieval with ColBERT and GraphRAG, and measure answer quality with Ragas. By the end, you have a working RAG engine you can explain step by step.
Fine-tuning LLMs and small language models
Sometimes a general model is not enough. In Module 4 you learn parameter-efficient fine-tuning with LoRA and QLoRA, fine-tune open models such as Qwen, Llama and Phi, and train a vision-language model to read documents and return clean, structured JSON. Module 5 then covers quantization and serving with vLLM and Ollama, so your models run fast and cheaply. Later, Module 8 introduces RLHF, DPO and GRPO, the methods used to make modern models reason better.
An Agentic AI course inside the program
Agentic AI is the next step after chatbots: AI systems that plan, use tools and complete tasks on their own. Module 6 teaches the ReAct framework, function calling, multi-agent workflows with LangGraph and CrewAI, and agent memory. Module 7 covers AI coding agents such as Claude Code, Gemini CLI and Aider, along with prompt engineering for real code bases and the Model Context Protocol. If you were searching for a separate Agentic AI course or a prompt engineering course in Pune, both are already part of this program.
Generative AI course with placement support
You are graded on a job-ready portfolio rather than written exams: weekly lab notebooks, a mid-term RAG engine, a fine-tuned document extractor and a final capstone product deployed with Docker and FastAPI. Alongside the course you get resume help, mock technical and HR interviews, and placement assistance from our team, with the option of an internship so your resume shows real project experience.
Online Generative AI course or classroom training in Pimpri Chinchwad
Join the classroom batch at our centre in Kunal Plaza, near Chinchwad Station, or attend a live online Generative AI course from anywhere. Both follow the same lectures, labs and trainers. You only need a laptop with 16GB RAM; free Google Colab covers the GPU work. With 8 batch slots between 9:30 AM and 8:30 PM, students and working professionals can both find a time that fits.
Generative AI course fees in Pune and certification
Fees for Generative AI courses in Pune vary widely, so check what is included before you decide. At Teknowell, the fee covers 36 live lectures, hands-on labs, portfolio projects, interview preparation, placement assistance, an ISO-certified course completion certificate and an internship certificate. Our team shares the exact fee on a quick call or WhatsApp message. The easiest way to decide is to book a free demo class and see a live lab for yourself.
Explore more
Related Courses and Guides
Data Science & Data Analytics
Data Science with Agentic AI (4 months) and Data Analytics with AI.
CourseCloud Computing & DevOps
AWS, Azure, GCP, Kubernetes and Terraform in 4 months.
CourseFull Stack Developer Course
Java with AI (6 months) or MEAN/MERN with AI (3 months).
GuideWill AI Replace Developers?
What AI means for software jobs in India.
Common queries
Frequently Asked Questions
Everything you need to know about the Teknowell Generative AI course.
Our counsellors reply on WhatsApp, usually within the hour during office hours.
Ask on WhatsAppCourse basics
What is the best Generative AI course in Pune for beginners?
A good Generative AI course for beginners teaches both how large language models (LLMs) work and how to build real applications with them. Teknowell's 3-month Generative AI course in Chinchwad, Pune does this with 36 live lectures, about 75% hands-on coding and 9 portfolio projects, starting from Python and neural network basics.
How long does it take to learn Generative AI (Gen AI)?
Most learners can become job-ready in Generative AI in about 3 months of structured practice. Teknowell's Generative AI course runs for 12 weeks, with 36 live lectures of 2 hours each (72 to 80 hours in total), plus weekly labs and projects.
Can I learn Generative AI without coding experience?
You can start the Teknowell Generative AI course with basic Python. The first lectures refresh Python, PyTorch and neural network basics, but the course is about 75% coding, so complete non-coders should first learn Python fundamentals or take a short Generative AI upskilling course.
Is Generative AI training available online in Pune?
Yes. Teknowell's Generative AI course is available as live online classes and as classroom training at the Chinchwad centre in Pune. Both options follow the same 36 lectures, labs and trainers.
Who can join the Generative AI course at Teknowell?
Graduates, current students, working professionals and candidates with an education gap can all join Teknowell's Generative AI course. Engineering students in Computer Science, IT, AI and Data Science get the most out of it, and basic Python is recommended.
Skills you learn
What is the difference between Generative AI and Agentic AI?
Generative AI creates content such as text, code or images in response to a prompt. Agentic AI goes a step further: an AI agent plans a task, uses tools and takes actions on its own to complete it. Teknowell's Generative AI course covers both, with AI agents taught in Modules 6 and 7.
Does the Generative AI course cover LLMs?
Yes. Teknowell's Generative AI course explains how large language models (LLMs) work, including tokenization, embeddings and self-attention, and students build a small NanoGPT-style transformer from scratch in Module 2 before working with open models such as Qwen, Llama and Phi.
What is RAG, and is it included in the course?
RAG (Retrieval-Augmented Generation) connects a language model to your own documents so it can answer questions from them accurately. It is a full module in Teknowell's Generative AI course, covering embeddings, vector databases such as Qdrant and Faiss, ColBERT retrieval, GraphRAG and evaluation with Ragas.
Will I learn to fine-tune LLMs?
Yes. In Teknowell's Generative AI course you fine-tune open models such as Qwen, Llama and Phi using LoRA and QLoRA, train a vision-language model to return structured JSON, and learn alignment methods such as RLHF, DPO and GRPO.
Is prompt engineering included in the Generative AI course?
Yes. Prompt engineering is taught in Module 7 of Teknowell's Generative AI course, including system instructions for real code bases, and it is practised throughout the RAG and AI agent modules.
Do I need a GPU laptop for the Generative AI course?
No. For Teknowell's Generative AI course, any laptop with 16GB RAM and VS Code is enough. Free-tier Google Colab with a T4 GPU covers all core sessions.
What projects will I build in the Generative AI course?
Students of Teknowell's Generative AI course build 9 projects: a tokenizer, a small transformer trained from scratch, a RAG engine, a fine-tuned model that outputs JSON, a fast model server, a multi-agent system, an AI-assisted coding workflow, a reasoning model tuned with DPO or GRPO, and a deployed capstone product.
Careers & placement
Is Generative AI a good career in India?
Generative AI is one of the fastest-growing areas in Indian IT, as services companies, product firms and startups build AI features into their products. Roles such as Generative AI Engineer, LLM Application Developer and AI Agent Developer are increasingly common in Pune's IT hubs, including Hinjawadi, Kharadi and Magarpatta.
What jobs can I get after a Generative AI course?
After Teknowell's Generative AI course, common roles include Generative AI Engineer, LLM Application Developer, AI Agent Developer, RAG Engineer, Prompt Engineer, AI/ML Engineer and LLMOps Engineer.
Does Teknowell provide placement assistance for the Generative AI course?
Yes. Teknowell provides placement assistance for Generative AI students, including resume building, mock technical and HR interviews, hiring connections and an internship option alongside training.
Should I choose Generative AI or Data Science?
Choose Generative AI if you want to build applications with LLMs, RAG and AI agents. Choose Data Science if you prefer data analysis, statistics and machine learning models. Teknowell offers both: the 3-month Generative AI course and the 4-month Data Science with Agentic AI course.
Fees, batches & location
What is the fee for the Generative AI course in Pune?
The fee for Teknowell's Generative AI course in Pune depends on the learning mode (classroom or online) and current batch offers, so the team shares the exact amount on a call or WhatsApp at +91 91460 11401. It includes 36 live lectures, labs, projects, interview preparation and certificates.
What are the batch timings for the Generative AI course?
Teknowell runs 8 Generative AI batch slots between 9:30 AM and 8:30 PM, including evening batches until 8:30 PM for working professionals.
Will I get a certificate after the Generative AI course?
Yes. Students who complete Teknowell's Generative AI course receive an ISO-certified course completion certificate, plus an internship certificate if they complete the internship.
Where is the Teknowell Generative AI institute located?
Teknowell EduTech is at Office No. 77, 5th Floor, Kunal Plaza, near Chinchwad Station, Old Mumbai Pune Highway, Chinchwad, Pune 411019, in Pimpri Chinchwad (PCMC). Phone: +91 91460 11401.
Free demo class
Book Your Free Demo Class
Generative AI course, in Chinchwad or live online. Leave your number and our team calls you with the next batch date.
Prefer to talk? Call +91 91460 11401 or WhatsApp us.