A 3-year deep-dive programme covering Generative AI, Agentic Systems, LLMs, RAG, Neural Networks, Data Science, and everything in between — with hands-on labs, real-world projects, and the industry tools you'll actually use on the job.
From Python fundamentals to building autonomous AI agents that rival research-grade systems.
Variables, control flow, functions, OOP, async/await, decorators, generators. Build a mini interpreter by week 3.
The mathematical bedrock of all ML. Interactive visualizations make it intuitive, not intimidating.
Fetching, cleaning, transforming, and storing data at scale. APIs, scrapers, SQL, and modern data stacks.
Supervised & unsupervised learning. Regression, classification, clustering — understand the why, not just the how.
Build a neural network using only NumPy. Understand every weight, bias, activation, and gradient.
Production-grade neural networks with TensorFlow. Cloud training, GPU acceleration, model serving.
Computer vision fundamentals. Object detection, image segmentation, real-time video analysis.
RNNs, LSTMs, GRUs — sequence-to-sequence learning. The stepping stone to understanding Transformers.
Tokenization, word embeddings, topic modelling, sentiment analysis — the NLP toolbox.
The architecture behind GPT, BERT, Claude, Gemini. Implement multi-head attention from first principles.
What LLMs are, how they're trained, how to talk to them via API. Prompt engineering essentials.
Synthesize everything learned. Build a full-stack dashboard that fetches live data, runs ML models, and generates LLM summaries.
How GPT-4, Claude 3, Llama 3, Mixtral were actually built. Data curation, tokenization, training at scale.
Adapt any LLM to your domain. LoRA, QLoRA, Prefix Tuning, Adapters — production fine-tuning on limited hardware.
Naive RAG → Advanced RAG → Modular RAG → GraphRAG. Build production knowledge systems.
Stable Diffusion, DALL-E, Flux, Sora internals. Train your own LoRA art style. Video generation.
ReAct, Plan-and-Execute, Reflexion, Tree-of-Thoughts. Build agents that think, act, and self-correct.
AutoGen, CrewAI, MetaGPT. Agents that delegate, collaborate, argue, and solve complex tasks together.
Agents that control computers. Playwright automation, GUI agents, Claude Computer Use, Operator.
Anthropic's open standard for connecting AI to any tool, database, or API. Build MCP servers from scratch.
Take models from notebook to production. CI/CD for ML, monitoring, A/B testing, drift detection.
Constitutional AI, RLHF alignment, red-teaming, jailbreaks, bias detection. The responsible AI practitioner.
Speech recognition, TTS, real-time voice agents. Vision-language models. Build a voice AI assistant.
Build a full agentic system: web browsing, document reading, RAG memory, report writing. Deployed and monitored in production.
Read and implement papers from DeepMind, Anthropic, OpenAI, Google Brain. Reproduce state-of-the-art results.
Build autonomous coding agents. Full software engineering pipelines: read issue → write code → run tests → submit PR.
Q-learning to PPO to GRPO. Train LLMs with RL. Build game-playing and robot control agents.
Train models across 100s of GPUs. FSDP, DeepSpeed, Megatron-LM. Cloud cost optimization.
Med-PaLM, BioGPT, AlphaFold. Clinical NLP, medical imaging, drug discovery. HIPAA-compliant systems.
ROS2, embodied AI, vision-language-action models. Connect LLMs to real hardware.
Sentiment trading, LLM financial analysts, fraud detection at scale. Backtesting frameworks.
12-week original research or industry-grade project. DARPA / Anthropic benchmark level. Peer-reviewed, published on GitHub, presented at Demo Day.
Interactive demos embedded in your learning environment. No setup required.
Every tool, framework, and platform used by AI engineers at Anthropic, Google, OpenAI, and the world's top AI labs.
From your first classifier to publishing AI research. Every project is deployable, demonstrable, and portfolio-ready.
Upload clinical notes, get differential diagnoses with cited medical literature. Vector search over PubMed + fine-tuned Bio-BERT.
Give it a GitHub issue URL. Watch it read code, write a fix, run tests, and submit a PR. Built with LangGraph + E2B sandboxing.
Stream Reddit + news, classify sentiment with fine-tuned FinBERT, generate trading signals. Live dashboard with Plotly.
Record a meeting → Whisper transcription → Claude summary → extract action items → push to Jira/Notion via MCP.
Upload product photo → remove background → generate 10 different scene styles with Flux/SD + ControlNet. SaaS-ready API.
Train a small language model to reason step-by-step using Group Relative Policy Optimization. Reproduce DeepSeek-R1-Zero methodology.
From zero to frontier AI researcher. Each skill unlocks the next.