Enrolling Now — 2025 Batch

Master the Full Stack of
Artificial Intelligence

India's most rigorous 3-year AI Diploma. From NumPy to Agentic AI systems — build real projects, ship production code, and graduate ready to work at Anthropic, Google DeepMind, or launch your own AI startup.

36
Months
72+
Real Projects
40+
Industry Tools
6
Capstones
∞
LMS Access
Explore Curriculum Access LMS

THE JOURNEY

Three Years. Full Mastery.

A carefully sequenced spiral curriculum where each year builds exponentially on the last — from data wrangling to deploying frontier AI agents.

YEAR 01 / MONTHS 1–12
Foundations & Deep Learning
  • Python for AI Engineers — NumPy, Pandas, async
  • Statistics, Probability & Linear Algebra
  • Scikit-learn — all classical ML algorithms
  • TensorFlow / Keras / PyTorch from scratch
  • CNNs, RNNs, Attention — build it all
  • 12 real-world ML projects with live datasets
🎯 Graduate as a Junior ML Engineer with production-ready GitHub
YEAR 02 / MONTHS 13–24
Generative AI & LLM Engineering
  • LLM architecture — reproduce GPT-2 from scratch
  • Anthropic Claude, GPT-4o, Gemini, Llama 3 APIs
  • RAG systems with vector databases
  • Fine-tuning with LoRA/QLoRA on Hugging Face
  • Multimodal AI — vision, audio, video
  • MLOps, production deployment, monitoring
🎯 Graduate as an LLM Engineer with deployed production apps
YEAR 03 / MONTHS 25–36
Agentic AI & Frontier Research
  • Agent design patterns — ReAct, MCP, tool-use
  • Multi-agent systems: CrewAI, AutoGen, LangGraph
  • Computer Use — desktop automation agents
  • Reproduce frontier research papers
  • 12-week industry-grade capstone project
  • Conference submissions & open-source contributions
🎯 Graduate as an AI Systems Engineer ready for Anthropic/DeepMind

SEMESTER BREAKDOWN

The Complete Lesson Plan

Every week mapped. Every project defined. Every tool introduced at exactly the right moment.

Semester 1: Data Science Mastery

Months 1–6 · Build unshakeable foundations in Python, Statistics, and Machine Learning. Every concept learned through real datasets and shipped projects.

6 Months 6 Projects Python · Pandas · Scikit-learn
MONTH 01
Python for AI Engineers
Foundation
W1
Python Mastery — Data Structures & Functional Patterns
List/dict comprehensions, generators, decorators, type hints, async/await. Interactive coding challenges in browser-embedded REPL.
W2
NumPy — Vectorization, Broadcasting & einsum
Stop writing for-loops. Learn to think in tensors. Matrix operations, FFT, random number generation. Visualized with interactive heatmaps.
W3
Pandas Mastery — Groupby, Merge, Time Series
Method chaining, MultiIndex, resample, rolling windows. Work with the NYC Taxi dataset (150M rows). Learn Polars for speed.
W4 🚀
PROJECT: Real-time Stock Market Dashboard
Fetch live data from Alpha Vantage API. Build interactive Plotly dashboard. Deploy to Streamlit Cloud. Ship it on GitHub.
Python 3.12 NumPy Pandas Polars Plotly Streamlit Alpha Vantage API
MONTH 02
Statistics & Probability for ML
Theory + Practice
W1
Probability Distributions & Bayes Theorem
Normal, Binomial, Poisson, Beta distributions. Interactive sliders to visualize parameter effects. Bayes theorem with real medical test examples.
W2
Hypothesis Testing & A/B Testing
t-tests, chi-squared, Mann-Whitney. p-values demystified. Design statistically valid A/B tests. The p-hacking problem. Multiple comparisons.
W3
Linear Algebra — Vectors, PCA, SVD
Eigenvalues, eigenvectors, matrix decomposition. PCA on real datasets. SVD for recommendation systems. Interactive 3D visualizations.
W4 🚀
PROJECT: A/B Test Analyzer for E-commerce
Complete statistical framework. Load real conversion data, run multiple tests, report with confidence intervals and power analysis. Used in real online stores.
SciPy StatsModels Matplotlib Seaborn PyMC
MONTH 03
Data Engineering & APIs
Engineering
W1
REST APIs, GraphQL & WebSockets
httpx, aiohttp, websocket-client. Build async data ingestion pipelines. Pagination, rate limiting, retry logic. OAuth flows.
W2
Data Cleaning & Feature Engineering
Outlier detection (IQR, isolation forest). Imputation strategies. Feature engineering: lag features, rolling stats, target encoding. Feature selection.
W3
SQL + NoSQL + DuckDB
Advanced SQL (window functions, CTEs, lateral joins). MongoDB aggregation pipeline. Redis caching. DuckDB for analytical queries at Python speed.
W4 🚀
PROJECT: Automated Multi-Source Data Pipeline
Ingest Weather (OpenMeteo) + News (NewsAPI) + Crypto (CoinGecko) data. Store in DuckDB. Auto-refresh every 15 min. Visual dashboard.
FastAPI httpx DuckDB PostgreSQL Redis Airflow
MONTH 04
Scikit-learn & Classical ML
Machine Learning
W1
Regression — From Linear to ElasticNet
OLS derivation by hand. Ridge, Lasso regularization intuition. ElasticNet. Polynomial regression. Interaction terms. Residual analysis.
W2
Classification — SVM, Trees, XGBoost
Decision boundaries visualized. Support vectors intuition. Random forests — feature importance. Gradient boosting from scratch. LightGBM vs XGBoost benchmarks.
W3
Clustering — K-Means, DBSCAN, Gaussian Mixture
Interactive cluster visualization. Elbow method and silhouette scores. DBSCAN for anomaly detection. GMM as soft clustering. Application: customer segmentation.
W4 🚀
PROJECT: Customer Churn Predictor (Kaggle Style)
Full ML pipeline: EDA → feature engineering → model selection → calibration → Shapley explanations → deployed FastAPI endpoint.
Scikit-learn XGBoost LightGBM SHAP Optuna FastAPI
MONTH 05
Advanced ML & Model Operations
MLOps
W1
Ensemble Methods — Stacking & Blending
Stacking with meta-learners. Blending vs stacking. Voting classifiers. How top Kaggle teams combine 50+ models. Build a competition-grade solution.
W2
Hyperparameter Tuning — Optuna & Ray Tune
Bayesian optimization vs grid search. Tree-structured Parzen Estimator (TPE). Pruning unpromising trials. Distributed HPO with Ray. 10x faster tuning.
W3
Model Evaluation — Beyond Accuracy
Calibration curves, Brier score. Class imbalance (SMOTE, class weights). Time-series cross-validation. Learning curves. Bias-variance tradeoff visualized.
W4 🚀
PROJECT: Full ML Pipeline with MLflow Tracking
Track 100+ experiments. Model registry. Automated retraining trigger. Data validation with Great Expectations. Deployed with BentoML.
MLflow Optuna Ray Tune BentoML DVC Great Expectations
MONTH 06
Data Visualization & Storytelling
CAPSTONE
W1
Matplotlib Mastery & Seaborn Statsplots
Custom figure composition. Subplots, twin axes. Seaborn pairplots, ridgeplots, violin plots. Publication-quality figures. Color theory for data.
W2
Plotly & Dash — Interactive Web Dashboards
Plotly Express for rapid charts. Dash callback architecture. Multipage apps. Mapbox integration. Real-time updates with dcc.Interval.
W3
D3.js for AI Visualizations
Build custom neural network visualizations. Force-directed graphs for knowledge graphs. Animated training progress. Integrate with Python backends.
W4 🏆
CAPSTONE S1: Kaggle Competition — Full Data Science Sprint
Real Kaggle competition entry. Full report with EDA, model iterations, and final submission. Present findings to cohort. Top 10% earns bonus XP.
Matplotlib Seaborn Plotly Dash D3.js Kaggle API

Semester 2: Deep Learning & Neural Networks

Months 7–12 · Build neural networks from pure math. Master TensorFlow, PyTorch, CNNs, RNNs, and the Transformer architecture. Every concept has working code.

6 Months 6 Projects TensorFlow · PyTorch · Keras
MONTH 07
Neural Networks from Scratch
Deep Learning Core
W1
Perceptrons & The Backpropagation Algorithm
Derive backprop by hand with chain rule. Compute gradients for a 3-layer network on paper before writing code. The "aha" moment of understanding gradient flow.
W2
Build a Neural Net in Pure NumPy
Forward pass, loss computation, backward pass, gradient descent — all in NumPy. No framework. This is the foundation that makes TF/PyTorch make sense.
W3
Activation Functions, Weight Init & Batch Norm
ReLU, GELU, Swish, SiLU — when to use each. He/Xavier initialization — why it matters. Batch norm, layer norm, group norm. Vanishing/exploding gradients.
W4 🚀
PROJECT: MNIST Classifier — Zero Frameworks
99%+ accuracy using only NumPy. Implement SGD, Adam, momentum from scratch. Visualize decision boundaries. Compare to sklearn's MLP.
NumPy Matplotlib Pure Python Math only
MONTH 08
TensorFlow & Keras Mastery
Framework
W1
TF Computation Graphs & @tf.function
Eager vs graph execution. @tf.function tracing, autograph. tf.GradientTape. Custom training loops vs model.fit(). TF Datasets API for efficient pipelines.
W2
Keras Functional API & Custom Layers
Multi-input/output models. Residual connections (ResNet style). Custom loss functions, metrics, callbacks. Mixed precision training (float16).
W3
TensorBoard, Checkpointing & TF Serving
Profile bottlenecks with TensorBoard. Model checkpointing strategies. Save/load models. TF Serving for production REST inference. TF Lite for mobile.
W4 🚀
PROJECT: Real-time Image Classifier (TF.js + Webcam)
Train on Google Colab TPU. Convert to TF.js. Deploy in browser — classify images live from webcam. Zero latency client-side inference.
TensorFlow 2.x Keras 3 TF.js TensorBoard Google Colab TPU
MONTH 09
Computer Vision
Vision AI
W1
CNNs — Convolutions, Pooling & Receptive Fields
Interactive conv filter visualization. Feature map evolution through layers. Receptive field calculation. Strided convolutions, dilated convolutions, depth-wise separable.
W2
Transfer Learning — ResNet, EfficientNet, ViT
ImageNet pre-training intuition. Feature extraction vs fine-tuning. Progressive unfreezing. Vision Transformer (ViT) patches. ConvNeXt vs ViT tradeoffs.
W3
Object Detection — YOLO v8, Segmentation
Anchor boxes, IoU, NMS. YOLO v8 training on custom dataset. SAM (Segment Anything Model). Instance vs semantic segmentation. Real-time inference optimization.
W4 🚀
PROJECT: Real-time Object Detection Web App
YOLO v8 trained on custom dataset. Deployed via FastAPI. Web frontend with live webcam feed. Detects objects + draws bounding boxes in real-time.
PyTorch Ultralytics YOLO torchvision OpenCV Segment Anything
MONTH 10
Sequence Models & NLP
NLP
W1
RNNs, LSTMs & GRUs — Theory + Code
Vanishing gradient problem animated. LSTM cell gates intuition. GRU simplification. Bidirectional RNNs. Sequence-to-sequence architectures. Language modeling.
W2
Attention Mechanism from Scratch
Bahdanau attention by hand. Query, Key, Value intuition. Scaled dot-product attention. Multi-head attention. Positional encodings. Attention weight visualization.
W3
Transformer Architecture — Encoder + Decoder
Build a mini Transformer from scratch in PyTorch. BERT (encoder only) vs GPT (decoder only) vs T5 (encoder-decoder). Tokenization, embeddings, layer norm placement.
W4 🚀
PROJECT: Real-time Twitter Sentiment Analyzer
Fine-tune BERT on Twitter data. Stream live tweets via API. Sentiment dashboard with real-time charts. Deployed with Streamlit + HuggingFace Spaces.
PyTorch Hugging Face Transformers Tweepy Streamlit
MONTH 11
PyTorch Deep Dive
Advanced
W1
PyTorch Internals — Autograd & Custom Ops
How autograd computes gradients. Writing custom autograd functions. torch.compile with TorchInductor. Triton kernel basics. CUDA extension tutorial.
W2
DataLoaders, Custom Datasets & Augmentation
Efficient data loading patterns. WebDataset for large-scale training. Albumentations for vision. torchtext/torchaudio. Multi-worker data loading profiling.
W3
Distributed Training & Mixed Precision
DDP vs FSDP vs DeepSpeed. AMP (automatic mixed precision). Gradient checkpointing. PyTorch Lightning for clean boilerplate. Train on 8-GPU setup (cloud).
W4 🚀
PROJECT: Custom BERT Fine-tuning for Domain NLP
Fine-tune BERT on domain-specific text (legal, medical, or code). Custom tokenizer. Task-specific heads. ONNX export for fast inference. Benchmark against GPT-3.5.
PyTorch 2.x PyTorch Lightning PEFT DeepSpeed ONNX Weights & Biases
MONTH 12
Intro to Generative Models
CAPSTONE S2
W1
Autoencoders & Variational Autoencoders
Encoder-decoder architecture. Latent space visualization. VAE reparameterization trick. ELBO loss. Interpolation between faces in latent space. CVAE.
W2
GANs — DCGAN, Conditional GAN
Adversarial training intuition. Mode collapse and how to avoid it. DCGAN architecture. Progressive GAN. StyleGAN2 concepts. Training stability tricks.
W3
Diffusion Models — DDPM Intuition
Forward diffusion process. Score matching intuition. DDPM vs DDIM. Classifier-free guidance. The noise prediction network (U-Net). Stable Diffusion overview.
W4 🏆
CAPSTONE S2: Generative Art Studio
Train a custom diffusion model on a curated dataset. Build web UI for generation. Deploy on Hugging Face Spaces. Present at cohort showcase.
Diffusers PyTorch Hugging Face Spaces Gradio DALL-E API

Semester 3: LLMs, RAG & Fine-tuning

Months 13–18 · Go deep on large language models. Build RAG pipelines. Fine-tune Llama 3. Ship production LLM applications with Claude & GPT-4o APIs.

6 Months LLM Engineering Claude · Llama · LangChain
MONTH 13
LLM Fundamentals
LLM Core
W1
Transformer Architecture Deep-Dive
Flash Attention. Rotary Position Embeddings (RoPE). Grouped Query Attention (GQA). KV cache. SwiGLU activation. ALiBi vs RoPE positional encoding tradeoffs.
W2
Tokenization — BPE, SentencePiece, tiktoken
Build a BPE tokenizer from scratch. Why tokenization matters for code/math. Token fertility across languages. tiktoken internals. Interactive tokenizer visualizer.
W3
Pre-training Objectives & Scaling Laws
MLM (BERT), CLM (GPT), prefix LM (T5). Chinchilla scaling laws. Compute-optimal training. Why bigger isn't always better. Data quality vs quantity.
W4 🚀
PROJECT: Reproduce GPT-2 (Karpathy's nanoGPT Style)
Train a 124M parameter GPT-2 from scratch on a custom text corpus. Full training loop, generation, and perplexity evaluation. Document every design decision.
PyTorch nanoGPT tiktoken W&B Google Colab A100
MONTH 14
LLM APIs & Prompt Engineering
APIs
W1
Anthropic Claude API — Streaming, Vision, Tools
Messages API, system prompts, streaming responses. Computer use API. Vision capabilities. Tool/function calling. Batch API for scale. Token counting and cost optimization.
W2
OpenAI, Gemini, Llama 3, Mistral APIs
Structured outputs with JSON mode. OpenAI Assistants API with file search. Gemini's 1M context window. Ollama for local models. Model routing strategies.
W3
Advanced Prompt Engineering
Chain-of-thought (CoT) reasoning. Few-shot vs zero-shot. Tree of Thought. Self-consistency. Prompt compression. Constitutional prompting. DSPy for automatic optimization.
W4 🚀
PROJECT: Multi-LLM Comparison Platform
Side-by-side comparison of 6 LLMs on custom benchmarks. Latency, cost, quality scoring. Automatic prompt optimization with DSPy. Interactive leaderboard.
Anthropic SDK OpenAI SDK Google Gemini Ollama DSPy LiteLLM
MONTH 15
RAG Systems Engineering
RAG
W1
Vector Databases — Pinecone, ChromaDB, pgvector
HNSW vs IVF-Flat indexing. Approximate nearest neighbor search. Metadata filtering. Hybrid search (dense + sparse). Qdrant, Weaviate, Milvus comparison. Build and benchmark all 5.
W2
Embedding Models & Semantic Search
text-embedding-3-large vs Cohere Embed v3 vs sentence-transformers. Matryoshka embeddings. Cross-encoder reranking. ColBERT late interaction. Embedding fine-tuning.
W3
Advanced RAG — Chunking, Reranking, HyDE
Recursive character splitting vs semantic chunking. Hypothetical Document Embeddings (HyDE). Multi-query retrieval. Contextual compression. RAG-Fusion. Self-RAG.
W4 🚀
PROJECT: Company Knowledge Base Chatbot
Ingest 500+ PDF documents. Smart chunking with metadata. Hybrid retrieval. Claude as the LLM. Conversation memory. Source citations. Deployed on Vercel.
Pinecone ChromaDB pgvector Cohere Rerank Claude API LangChain
MONTH 16
LLM Fine-tuning
Fine-tuning
W1
Full Fine-tuning, LoRA & QLoRA Theory
When to fine-tune vs RAG vs prompting. LoRA math: low-rank decomposition intuition. QLoRA: 4-bit quantization + LoRA. PEFT library internals. Adapter layers.
W2
Hugging Face PEFT & Training Pipeline
trl (Transformer Reinforcement Learning) library. SFTTrainer for supervised fine-tuning. Dataset preparation, formatting. Chat templates. Unsloth for 2x faster training.
W3
RLHF, DPO & Constitutional AI
RLHF pipeline: SFT → reward model → PPO. Direct Preference Optimization (DPO) — simpler than RLHF. Constitutional AI from Anthropic. RLAIF. Orpo, SimPO.
W4 🚀
PROJECT: Fine-tune Llama 3 on Custom Dataset
Prepare domain-specific instruction dataset (1000+ examples). QLoRA fine-tune on A100. Push to Hugging Face Hub. Benchmark against base model with LLM-as-judge.
PEFT trl Unsloth bitsandbytes Hugging Face Hub Modal
MONTH 17
LangChain, LlamaIndex & LangGraph
Orchestration
W1
LangChain — Chains, Agents, Memory, Tools
LCEL (LangChain Expression Language). Runnable interfaces. Conversation buffer, summary, and entity memory. Tool integration (search, code execution, databases).
W2
LlamaIndex — Document Processing & Query Engines
Document loaders (PDF, Notion, GitHub). Index types: VectorStore, Summary, Knowledge Graph. SubQuestion QueryEngine. RouterQueryEngine. Streaming responses.
W3
LangGraph — Stateful Agent Workflows
Graph-based agent design. State machines for LLMs. Conditional edges. Human-in-the-loop checkpoints. Persistence with SQLite/Postgres. Streaming partial outputs.
W4 🚀
PROJECT: Full-Stack RAG App with LangChain
Next.js frontend + FastAPI backend. LangGraph agent with tools (web search, code exec, SQL). Streaming response UI. Auth, rate limiting, usage analytics.
LangChain LlamaIndex LangGraph LangSmith Next.js FastAPI
MONTH 18
Production LLM Systems
CAPSTONE S3
W1
Safety, Guardrails & Prompt Injection Defense
Nemo Guardrails. Rebuff prompt injection detector. Input/output validation. PII detection and redaction. Content filtering strategies. Red-teaming your own LLM apps.
W2
LLM Evaluation — RAGAS, BLEU, LLM-as-Judge
RAGAS for RAG evaluation (faithfulness, relevance, context precision). MT-Bench style evaluation. LLM-as-judge with GPT-4o. Building evaluation harnesses. MMLU-style benchmarks.
W3
Semantic Caching, Latency & Cost Optimization
GPTCache semantic caching. Prompt compression (LLMLingua). Speculative decoding. Model cascading (cheap → expensive). Token budgeting. 10x cost reduction strategies.
W4 🏆
CAPSTONE S3: Production RAG Platform
Multi-tenant RAG platform. Document ingestion pipeline. Evaluation dashboard. Cost monitoring. Rate limiting. Deployed on AWS. 100+ concurrent users tested.
RAGAS Nemo Guardrails GPTCache LangSmith AWS Prometheus

Semester 4: Multimodal AI & Production Systems

Months 19–24 · Vision-language models, audio AI, diffusion models, and mastering the entire MLOps stack for production deployment at scale.

6 Months Multimodal MLOps · Cloud · Production
MONTH 19
Vision-Language Models
Multimodal
W1
CLIP, BLIP-2 & LLaVA Architecture
Contrastive learning in CLIP. Image-text alignment. BLIP-2 Q-Former bridge. LLaVA visual instruction tuning. InternVL2, Pixtral, Qwen-VL comparison.
W2
Claude Vision & GPT-4o Vision APIs
Image analysis, PDF understanding, chart reading. Multi-image reasoning. Vision grounding. Document OCR via LLMs. Processing entire codebases via images.
W3
Visual Q&A, Document AI & OCR
DocVQA, ChartQA, TextVQA benchmarks. Structured data extraction from PDFs. Invoice parsing. Receipt understanding. PaddleOCR + LLM pipeline.
W4 🚀
PROJECT: Visual Q&A App with PDF Analysis
Upload any PDF or image. Ask questions in natural language. Get accurate answers with highlighted source regions. Multi-document comparison mode.
Claude APILLaVACLIPPaddleOCRGradio
MONTH 20
Audio & Video AI
Audio/Video
W1
Whisper ASR & Speech Recognition Pipelines
Whisper.cpp for local inference. Speaker diarization with pyannote. Real-time transcription. Word-level timestamps. Language detection. WhisperX improvements.
W2
Text-to-Speech — ElevenLabs, Bark, XTTS
Voice cloning with 5 seconds of audio. Emotion control. Multi-language TTS. XTTS-v2 for open-source voice cloning. Real-time voice synthesis latency optimization.
W3
Video Understanding & Action Recognition
Video captioning with CogVLM2-Video. Action recognition with SlowFast. Video search with CLIP. Optical flow. Dense video captioning for long videos.
W4 🚀
PROJECT: AI Podcast Summarizer + Translator
Upload any podcast. Whisper transcription + speaker labels. Claude summarization. Multi-language audio output (ElevenLabs). Chapter generation. Share as newsletter.
WhisperpyannoteElevenLabsXTTS-v2FFmpeg
MONTH 21
Diffusion Models & Image Generation
GenAI
W1
Diffusion Theory — DDPM, DDIM, Score Matching
Forward diffusion as adding Gaussian noise. Reverse process as denoising. Score function estimation. DDIM deterministic sampling. Flow matching (simpler foundation).
W2
Stable Diffusion, ControlNet & IP-Adapter
Latent diffusion models. SDXL architecture. ControlNet for pose/depth conditioning. IP-Adapter for style transfer. Inpainting and outpainting. Textual inversion.
W3
ComfyUI Workflows, Flux & SDXL-Lightning
ComfyUI node-based generation. Flux.1 (Black Forest Labs). SDXL-Lightning 4-step generation. Video generation with AnimateDiff. Image editing with InstructPix2Pix.
W4 🚀
PROJECT: AI Creative Studio Web App
Text-to-image, image-to-image, inpainting, face swap, style transfer — all in one Gradio app. ControlNet integration. Gallery with prompt history. Deployed on Replicate.
DiffusersComfyUIControlNetFlux.1Replicate
MONTH 22
MLOps & Production Deployment
DevOps
W1
Docker, Kubernetes & FastAPI for ML
Containerize ML models. Multi-stage Dockerfiles. Kubernetes HPA for auto-scaling. Helm charts. Istio service mesh basics. GPU scheduling on K8s. NVIDIA container toolkit.
W2
AWS SageMaker & GCP Vertex AI
SageMaker training jobs, endpoints, pipelines. Vertex AI custom training, model registry. Cloud TPU training. Cost optimization (spot instances). Auto-scaling strategies.
W3
CI/CD for ML — GitHub Actions, DVC, ZenML
Automated testing for ML pipelines. Data versioning with DVC. Model versioning. ZenML stack composition. Automated retraining triggers. Model performance regression detection.
W4 🚀
PROJECT: Production ML API with Monitoring
FastAPI inference server. Prometheus metrics. Grafana dashboards. Alerting on drift. A/B testing two model versions. Auto-rollback on performance drop.
DockerKubernetesAWS SageMakerZenMLPrometheusGrafana
MONTH 23
AI System Design & Inference Optimization
Performance
W1
Model Compression — Quantization & Pruning
INT8/INT4/FP8 quantization. GPTQ, AWQ, GGUF formats. Structured vs unstructured pruning. Knowledge distillation. 4x model size reduction with <2% accuracy loss.
W2
Inference Optimization — vLLM & TensorRT-LLM
PagedAttention in vLLM. Continuous batching. Speculative decoding. TensorRT-LLM for NVIDIA GPUs. Medusa multi-token prediction. 10x throughput improvements.
W3
Cost Optimization & Batch Inference
Anthropic Batch API (50% cost reduction). Async inference strategies. Model cascading architecture. Cold start optimization for serverless. Token budget management.
W4 🚀
PROJECT: High-Performance Inference Server
vLLM serving Llama 3 at 1000+ tokens/sec. Dynamic batching. Load balancer with health checks. Benchmark vs naive PyTorch. Document the 20x speedup.
vLLMTensorRT-LLMGPTQAWQTritonlocust
MONTH 24
AI Ethics, Safety & Responsible AI
CAPSTONE S4
W1
AI Bias, Fairness Metrics & Audit Frameworks
Demographic parity, equalized odds, calibration. Fairlearn library. Algorithmic auditing. Counterfactual fairness. Disparate impact analysis. Building bias dashboards.
W2
Constitutional AI & Red-teaming
Anthropic's Constitutional AI paper. RLHF vs CAI. Red-teaming methodologies. Adversarial prompting. Jailbreak taxonomy. Building AI safety evaluation harnesses.
W3
EU AI Act & AI Safety Research
EU AI Act risk categories. Transparency requirements. High-risk AI system compliance. Interpretability (mechanistic interpretability primer). Alignment research overview.
W4 🏆
CAPSTONE S4: Multimodal AI Platform
End-to-end multimodal app: text + image + audio + video. Full safety layer. Bias monitoring. Usage dashboard. EU AI Act compliance checklist. Public demo day.
FairlearnResponsible AI ToolboxLLM GuardClaudeWeights & Biases

Semester 5: Agentic AI & Multi-Agent Systems

Months 25–30 · The frontier of AI. Build autonomous agents with memory, planning, tool use. Design multi-agent systems. Reproduce research papers.

6 Months Agentic AI MCP · CrewAI · LangGraph
MONTH 25
Agent Fundamentals
Agents
W1
ReAct Pattern & Tool Use
Reason + Act loop. Function calling internals. Tool selection strategies. Parallel tool calls. Error recovery patterns. Chain-of-tool-use. OpenAI parallel functions.
W2
Memory Systems — Episodic, Semantic, Working
In-context memory (limited). External memory with vector DBs. Procedural memory (fine-tuned behaviors). Memory consolidation. MemGPT architecture. Zep memory layer.
W3
Planning — MCTS, BFS & Task Decomposition
Plan-and-execute agents. BabyAGI planning. Tree-of-Thought planning. MCTS with LLMs. Hierarchical task decomposition. ReWOO (reduce observatIOns) for efficiency.
W4 🚀
PROJECT: Personal AI Research Agent
Given any research topic: searches arXiv + web, reads papers, synthesizes findings, generates a structured report with citations. Memory across sessions. Slack notifications.
Claude APILangGraphZeparXiv APITavily Search
MONTH 26
Advanced Agent Frameworks
MCP
W1
Anthropic Model Context Protocol (MCP)
MCP architecture: hosts, clients, servers. Build custom MCP servers. Resources, tools, prompts primitives. MCP vs OpenAI plugin system. Integrate with Claude Desktop.
W2
OpenAI Assistants API & Code Interpreter
Assistants v2 with file search + code interpreter. Thread management. Streaming with Server-Sent Events. Building a persistent AI assistant with function calling and file analysis.
W3
Browser Agents — Playwright + LLMs
Browser Use library. DOM understanding for LLMs. Screenshot-based navigation. Playwright automation with LLM decision making. Anti-detection patterns. Captcha handling.
W4 🚀
PROJECT: Automated Web Research Agent
Given a question: autonomously browses web, fills forms, extracts data, verifies information across sources. Produces sourced report. Deployed as API service.
Anthropic MCPBrowser UsePlaywrightPuppeteerAgentOps
MONTH 27
Multi-Agent Systems
Multi-Agent
W1
AutoGen, CrewAI & MetaGPT
AutoGen conversable agents. CrewAI role-based teams. MetaGPT software company simulation. Mixture-of-Agents for consensus. Agent communication protocols.
W2
Agent Communication & Consensus
Message passing architectures. Shared blackboard patterns. Voting mechanisms. Debate-and-refine. Society of Mind patterns. Emergence in multi-agent systems.
W3
Competitive vs Cooperative Multi-Agent RL
MARL (Multi-Agent RL) overview. Cooperative: QMIX, MAPPO. Competitive: MADDPG. Self-play training. OpenSpiel framework. LLM-based MARL hybrids.
W4 🚀
PROJECT: AI Software Engineering Team
Multi-agent system: PM agent → architect → developer → QA → DevOps. Given a spec, they autonomously build and deploy a web app. MetaGPT + GitHub integration.
CrewAIAutoGenMetaGPTLangGraphAgentOps
MONTH 28
Computer Use & Desktop Automation
Computer Use
W1
Claude Computer Use API
Computer use tool: bash, text editor, browser. Screenshot-action loop. Error recovery. Sandboxed Docker environments. Building reliable computer use pipelines.
W2
GUI Agents & Screen Understanding
UI-TARS architecture. SeeAct framework. ShowUI. Screen2Words. Building GUI understanding datasets. Accessibility tree parsing. Cross-platform automation.
W3
RPA + AI Integration
UiPath + LLM integration. Power Automate AI Builder. n8n workflow automation with AI nodes. Document processing automation. Invoice-to-database pipeline.
W4 🚀
PROJECT: AI Desktop Automation Suite
Automate: email triage, calendar management, spreadsheet analysis, report generation. Claude Computer Use + custom tools. Full audit trail. Error recovery.
Claude Computer UsePyAutoGUIn8nDockerxdotool
MONTH 29
Frontier Research Topics
Research
W1
World Models & Model-Based RL
Dreamer v3, RSSM. Learning world models from video. Planning in latent space. Curiosity-driven exploration. DIAMOND (diffusion world model). Genie 2 overview.
W2
Mixture of Experts (MoE) Architectures
Sparse MoE vs dense transformers. Mixtral architecture. Switch Transformer. Expert routing analysis. Load balancing losses. MoE for faster inference. Grok-1 analysis.
W3
Constitutional AI & Scalable Oversight
Anthropic's Constitutional AI paper deep-dive. Debate as alignment method. Recursive reward modeling. Weak-to-strong generalization. IRL approaches. AUP (Assistance Games).
W4 🚀
PROJECT: Reproduce a 2025 Research Paper
Pick from list of recent arXiv papers. Full reproduction: data, training, evaluation. Write a blog post on findings. Submit to ML Reproducibility Challenge.
PyTorchJAX/FlaxarXivPapers With CodeW&B
MONTH 30
Pre-Capstone Sprint
Career Prep
W1
AI System Design Interviews
Design: YouTube Recommendations, Fraud Detection System, LLM API Platform, Real-time Bidding. Practice with timing. Common mistakes at Anthropic/Google interviews.
W2
Portfolio Optimization — GitHub & Hugging Face
Craft README files that get stars. Demo GIFs for every project. Hugging Face model cards. Kaggle profile optimization. Open-source contribution strategy.
W3
Mock Technical Interviews (x5)
5 live mock interviews with industry mentors. ML fundamentals, coding, system design, research. Recorded for self-review. Detailed written feedback.
W4 🏆
CAPSTONE KICKOFF: 12-Week Industry Project
Project selection, team formation, mentor assignment, problem definition, initial research. Full spec document submitted and approved before Month 31.
GitHubHugging FaceKaggleLinkedInPortfolio site

Semester 6: Capstone, Industry & Graduation

Months 31–36 · Build your industry-grade capstone. Contribute to open-source. Nail your interviews. Graduate at the Demo Day with job offers in hand.

6 Months Capstone Industry-Ready
MONTHS 31–33
Industry-Grade Capstone Project (12 Weeks)
MAIN PROJECT
Track A
🏥 Healthcare AI — Diagnostic Assistant
Multi-modal diagnosis (X-ray + clinical notes + lab values). Explainability with SHAP/LIME. HIPAA compliance framework. FDA AI/ML guidance compliance. Real hospital dataset partnership.
Track B
💹 FinTech AI — Fraud Detection + Trading Agent
Real-time fraud detection (99.9% precision). Autonomous trading agent with backtesting. Risk management engine. SEC/SEBI regulatory compliance. Live paper trading deployment.
Track C
⚖️ Legal AI — Contract Analysis + Case Research
RAG over 100K legal documents. Contract clause extraction + risk flagging. Case law precedent search. Multi-jurisdiction support. Integration with legal software.
Track D
🎓 EdTech AI — Adaptive Learning Platform
Student knowledge modeling (BKT/DKT). Personalized question generation. AI tutor with Socratic method. Learning analytics dashboard. Deployed for real students.
All Year 1-3 ToolsClaudeAWS/GCPFull MLOps Stack
MONTH 34
Open Source & Research Contributions
Community
OS1
Open-Source Contribution Sprint
Contribute to: Hugging Face Transformers, LangChain, or CrewAI. Minimum: 1 merged PR. Document the contribution process. Build relationship with maintainers.
OS2
Conference Paper Submission
Submit to NeurIPS/ICLR workshops, EMNLP, or top Indian AI conferences. Full paper writing with related work, methodology, experiments, ablation studies.
OS3
AI Blog & Technical Writing
Write 3 technical blog posts explaining your capstone work. Publish on Medium, Substack, or personal site. LinkedIn thought leadership. Grow to 1000+ followers.
GitHubHugging FaceOverleaf LaTeXSubstack
MONTH 35
Career Accelerator
Career
C1
Resume, LinkedIn & Portfolio Polish
AI-industry resume format. Quantify every achievement. LinkedIn optimization for recruiter search. Personal portfolio site with live demos. Hugging Face & Kaggle profile review.
C2
10 Live Technical Interview Simulations
Coding (LeetCode ML problems), ML fundamentals, system design, behavioral. Mock interviews with industry mentors from top AI companies. Recorded + transcribed for review.
C3
Salary Negotiation & Startup Equity
Market rates for AI roles in India, UK, US, UAE. Negotiation tactics. Equity 101 — vesting, cliff, dilution. ESOP valuation. When to take a startup offer vs FAANG.
Interview prepGlassdoorlevels.fyiLinkedIn Premium
MONTH 36
Graduation & Demo Day
🎓 GRADUATION
G1
Public Demo Day — Industry Judges
10-minute demo + 5-minute Q&A. Live demo of capstone. Industry judges from Anthropic, Google, Indian AI startups. Best projects win internship offers.
G2
Portfolio on Hugging Face Spaces
All 6 capstone projects live on HF Spaces with interactive demos. Permanent showcase of 3-year learning journey. Public portfolio for recruiters.
G3 🎓
GRADUATION: Remesys-Tech Diploma in AI
Official diploma ceremony. Alumni network access. Lifetime LMS access (learn.remesys.in). Placement assistance. Welcome to the Remesys AI alumni community.
Hugging Face SpacesGitHubLinkedInlearn.remesys.in

TOOLS ECOSYSTEM

40+ Industry Tools

Every tool you'll use on day one at Anthropic, Google DeepMind, or your own AI startup. Introduced in context — when you actually need them.

🧮 Core ML / DL

  • TensorFlow 2.x + Keras 3
  • PyTorch 2.x + Lightning
  • Scikit-learn + XGBoost
  • JAX + Flax (intro)
  • Hugging Face Transformers
  • Diffusers library

🤖 LLM & GenAI

  • Anthropic Claude (all models)
  • OpenAI GPT-4o + Assistants
  • Google Gemini Pro
  • Llama 3, Mistral, Phi-4
  • Ollama (local inference)
  • LangChain + LangGraph
  • LlamaIndex
  • DSPy

🗄️ Vector & Data

  • Pinecone, ChromaDB, Qdrant
  • pgvector (PostgreSQL)
  • Elasticsearch
  • Pandas + Polars
  • DuckDB
  • Apache Spark (PySpark)

⚙️ MLOps & Cloud

  • MLflow + Weights & Biases
  • DVC (data versioning)
  • ZenML pipelines
  • Docker + Kubernetes
  • AWS SageMaker
  • GCP Vertex AI
  • Hugging Face Hub
  • Modal, Replicate

🕵️ Agentic

  • Anthropic MCP SDK
  • CrewAI, AutoGen
  • MetaGPT
  • Browser Use + Playwright
  • AgentOps monitoring
  • LangSmith tracing

🚀 Inference & Serving

  • vLLM (PagedAttention)
  • TensorRT-LLM (NVIDIA)
  • ONNX Runtime
  • FastAPI + BentoML
  • Triton Inference Server
  • Ollama (local)

REAL WORLD PROJECTS

72+ Projects, All Shipped.

Not toy datasets. Not contrived examples. Real projects that go on your GitHub and get you hired.

Year 1
📊

Real-time Stock Dashboard

Live market data from Alpha Vantage API. Pandas analysis pipeline. Plotly interactive charts. Deployed on Streamlit Cloud. Technical indicators overlay.

PythonPandasPlotlyAPI
Year 1
🔬

Customer Churn Predictor

Full ML pipeline with XGBoost + SHAP explainability. FastAPI endpoint. Docker deployment. 94% AUC on real telecom dataset. Used by real businesses.

XGBoostSHAPFastAPIDocker
Year 1
🧠

MNIST Neural Net (NumPy Only)

99.2% accuracy using zero ML frameworks. Pure NumPy backpropagation. Adam optimizer from scratch. Proves deep understanding of gradient descent math.

NumPyMathBackprop
Year 1
📡

Real-time Object Detection App

YOLO v8 on custom 5K-image dataset. Live webcam inference. FastAPI backend. Web frontend. 45+ FPS on consumer GPU. Bounding box visualization.

YOLO v8PyTorchFastAPIWebRTC
Year 2
⚡

GPT-2 Reproduction

Full 124M parameter GPT-2 trained on custom corpus. nanoGPT architecture. 40GB of text data. Reproduces original paper results within 2%. Full write-up.

PyTorchA100 GPUtiktokenW&B
Year 2
🔍

Production RAG Platform

Multi-tenant knowledge base platform. 500+ PDF ingestion. Hybrid search. Claude-powered answers with citations. 100+ concurrent users tested. AWS deployed.

Claude APIPineconeLangChainAWS
Year 2
🎨

AI Creative Studio

Text-to-image, inpainting, style transfer, ControlNet — all in one web app. ComfyUI backend + React frontend. Prompt history gallery. Deployed on Replicate.

Stable DiffusionControlNetFlux.1React
Year 3
🤖

AI Software Engineering Team

Multi-agent system: PM → Architect → Developer → QA. Given a spec, autonomously builds and deploys working web apps. MetaGPT + GitHub Actions integration.

MetaGPTCrewAILangGraphGitHub
Year 3
🖥️

AI Desktop Automation Suite

Claude Computer Use API. Automates email triage, calendar scheduling, spreadsheet analysis, report generation. Full audit trail. Enterprise-ready architecture.

Claude CU APIDockern8nPython

YEAR 3 CAPSTONE

DARPA-Level Final Projects

12-week industry-grade capstones that solve real problems. These aren't assignments — they're products.

🏥 Healthcare AI Diagnostic Assistant

Multi-modal diagnostic AI combining chest X-rays, clinical notes, and lab values. Full explainability with SHAP. HIPAA compliance framework. Real hospital dataset (IRB approved).

Vision TransformerClaude APISHAPFastAPI
Difficulty: ★★★★★ · MIT Med School difficulty

💹 FinTech Fraud Detection + Trading Agent

Real-time fraud detection at 99.9% precision. Autonomous trading agent with backtesting on 5 years of market data. Full SEBI compliance documentation.

Graph Neural NetRL AgentKafkaPostgreSQL
Difficulty: ★★★★★ · Goldman Sachs difficulty

⚖️ Legal AI Contract Intelligence Platform

RAG over 100K legal documents. Contract clause risk scoring. Case precedent search across jurisdictions. Paralegal-level accuracy. Partnership with legal firm.

RAGLegal NERClaudepgvector
Difficulty: ★★★★☆ · LexisNexis difficulty

🎓 Adaptive AI Learning Platform

Student knowledge tracing with Deep Knowledge Tracing (DKT). AI tutor using Socratic method. Personalized question generation. Deployed for real students — yours is the Remesys platform.

DKT ModelClaude TutorReactPostgreSQL
Difficulty: ★★★★☆ · Khan Academy difficulty

🤖 Open-Source AI Research Agent

Fully autonomous research assistant that reads papers, runs experiments, and generates novel hypotheses. Publishes findings. MCP-integrated with all major research tools.

LangGraphMCPClaudearXiv API
Difficulty: ★★★★★ · Anthropic difficulty

HOW YOU'RE EVALUATED

Portfolio-First Assessment

No memorization tests. You're evaluated on shipped code, Kaggle rankings, and paper reproductions — exactly like industry.

30%
Weekly Projects
Every week ships a project to GitHub. Graded on code quality, documentation, and working demo.
20%
Monthly Capstones
6 major capstones across 3 years. Peer-reviewed and judged by industry mentors via live demo.
20%
Competitions
Kaggle rankings, hackathons, internal leaderboards. Real competitive performance metrics.
10%
Paper Reproductions
Reproduce 1 research paper per month. Forces deep understanding. Submitted to ML Reproducibility Challenge.
20%
Final Capstone
The 12-week industry project. Public Demo Day with industry judges. This is your graduation piece.

GAMIFIED LEARNING

Earn XP. Level Up.

Learning shouldn't feel like work. Every achievement earns XP and unlocks badges — just like your favorite game, but the reward is a career.

🚀
First Deploy
Ship your first project to production
+500 XP
📊
Kaggle Knight
Reach top 20% in a Kaggle competition
+2000 XP
⭐
GitHub Star
Get 100 stars on a project
+3000 XP
📄
Paper Reproducer
Successfully reproduce an arXiv paper
+5000 XP
🤝
Open Source
Get a PR merged to a major repo
+4000 XP
🎤
Presenter
Present at a tech meetup or conference
+3500 XP
🧠
Neural Net Monk
Build a net from scratch (no frameworks)
+2500 XP
🏆
Capstone Champion
Win the semester capstone showcase
+10000 XP
2025 Batch — Limited Seats

Start Your AI Journey

Contact for Pricing

EMI options available · Full scholarship for exceptional students

  • Full 36-month curriculum with live sessions
  • Lifetime access to learn.remesys.in LMS
  • 1-on-1 mentorship from industry practitioners
  • GPU cloud credits for all projects
  • API credits: Claude, OpenAI, Pinecone, AWS
  • Career placement assistance
  • Alumni network access
  • Internship placement priority at partner companies
Apply Now → Explore LMS

LEARN BY DOING

Built-In Interactive Labs

Every concept comes with a playable simulation. No boring slides — just tools you can break, tweak, and understand.

LAB 01 · YEAR 1 · MONTH 7

Neural Network Visualizer

Watch signals flow through layers in real-time. Adjust architecture, activate neurons, see backprop gradients.

Loss: 0.4231
Epoch: 0
Params: 0
Status: Ready
LAB 02 · YEAR 1 · MONTH 2

Tensor Operations Playground

Click cells to edit values. See matrix multiply, transpose, and broadcasting happen live. Color = magnitude.

Matrix A (3×3)
Matrix B (3×3)
Result (A⊗B)
LAB 03 · YEAR 2 · MONTH 14

Prompt Engineering Lab

Compare Zero-shot vs Few-shot vs Chain-of-Thought prompting side by side. See how prompt structure changes LLM reasoning.

ZERO-SHOT
"Is this review positive or negative?
Review: The food was cold and service slow."
→ Model output:
"Negative"
Accuracy: ~78% on complex cases
FEW-SHOT
"Positive: Great food! · Negative: Terrible. · Positive: Amazing! ·
Review: Cold food, slow service →"
→ Model output:
"Negative — poor quality, slow service"
Accuracy: ~91% on complex cases
CHAIN-OF-THOUGHT
"...Let me think step by step. Cold food = negative signal. Slow service = negative signal. Therefore:"
→ Model output:
"Strongly Negative (2/2 signals)"
Accuracy: ~97% on complex cases
🔬 YOUR TURN — Edit and run prompts on learn.remesys.in
Classify the sentiment. Think step by step. Review: "The movie had stunning visuals but the plot made no sense."
LAB 04 · YEAR 2 · MONTH 15

RAG Pipeline Simulator

Watch how a query flows through a Retrieval-Augmented Generation system — from embedding to retrieval to LLM answer.

LAB 05 · YEAR 1 · MONTH 10

Attention Weight Visualizer

Click any token to see its attention distribution. Line thickness = attention weight. This is how Transformers "read" text.

Click a token to see its attention. Selected: none

36-MONTH JOURNEY

Your Learning Path

Every skill builds on the last. By Month 36 you've touched every layer of the AI stack — from data wrangling to frontier agent systems.

WEEKLY RHYTHM

A Typical Week

Structured enough to keep you on track. Flexible enough to go deep when something clicks.

MONDAY
Theory + Concept
Live session (2h). Instructor explains the week's core concept with interactive demos. Q&A.
TUESDAY
Hands-on Lab
Guided coding lab (3h). Follow along with LMS notebook. Solve 5 challenges. Stuck? AI tutor helps.
WEDNESDAY
Paper Reading
Read 1 arXiv paper (1h). Post a 200-word summary to cohort. Discuss in Slack. Build citation habit.
THURSDAY
Project Sprint
Work on the weekly project (4h). Build it from scratch. Use all tools. Ship to GitHub by Friday.
FRIDAY
Demo & Critique
"AI Roast" (1h). 3 students demo their project. Cohort gives feedback. Best project earns XP bonus.
WEEKEND
Self-directed
Optional deep dives, Kaggle competitions, open-source contributions. No mandatory sessions.

HOW WE COMPARE

MIT/Harvard Level Curriculum

We benchmarked our curriculum against the world's best. Here's how we stack up — and where we go further.

TOPIC MIT 6.S191 CS229 Stanford Fast.ai REMESYS ★
Deep Learning Foundations ✓ ✓ ✓ ✓ + From Scratch
LLM Engineering & APIs Partial Partial ✗ ✓ Full Semester
RAG Systems ✗ ✗ ✗ ✓ + Production
Fine-tuning (LoRA/QLoRA) ✗ ✗ Partial ✓ + RLHF/DPO
Agentic AI Systems ✗ ✗ ✗ ✓ Full Semester
MLOps & Production Partial ✗ ✓ ✓ Cloud + K8s
Multimodal AI ✓ Partial ✗ Partial ✓ Text+Image+Audio+Video
Paper Reproductions ✗ Some ✗ ✓ Monthly
Placement / Career Track ✗ ✗ ✗ ✓ Full Semester

FREQUENTLY ASKED

Got Questions?

Do I need prior programming experience?
Basic Python helps but isn't required. Month 1 covers everything you need. We've successfully graduated students from 0 coding background.
What hardware do I need?
Just a laptop with internet. All GPU compute is provided via Google Colab Pro + AWS/GCP credits included in the program. No expensive GPU required.
How many hours per week?
15–20 hours/week for full-time engagement. Live sessions are recorded for async watching. Weekend sessions are optional. Designed for working professionals too.
Will I get API credits?
Yes. Program includes credits for: Anthropic Claude, OpenAI, Pinecone, AWS, GCP, Replicate. Enough to complete all projects without extra spend.
What's the placement rate?
Our graduates have placed at Infosys AI, TCS Innovation, Google India, startups, and international remote roles. Portfolio quality is the primary placement driver.
Is learn.remesys.in included?
Yes. Lifetime access to the LMS at learn.remesys.in including all notebooks, datasets, recorded sessions, and future curriculum updates at no extra cost.