Official website for Dr. TMA Pai Endowment Chair - ITIS.
This website showcases our open documentation in the field of Information Technology and Intelligent Systems.
We are committed to advancing knowledge and innovation in technology and education.
Models & Training18
J-Space: A Verbalizable Workspace Inside Language Models
Multi-Token Prediction: Training Draft Heads for Faster Decoding
Scaling Laws: What the Curves Predict and What They Hide
Model Compression Beyond Quantization: Pruning, Distillation, and Low-Rank Methods
Model Alignment Is a Control Problem
LLM Training Optimizers: AdamW, Adafactor, Shampoo, and Muon
Frontier Mechanistic Interpretability: From Features to Causal Audits
Diffusion and Flow Matching: Architectures for Image and Video Generation
Long-Context Models: Position Encoding, Memory, and Retrieval
Synthetic Data Pipelines: Generation, Filtering, and Validation
LLM Evaluation: Contamination, Judge Bias, and Production Metrics
Distributed LLM Training: Data, Tensor, Pipeline, and Expert Parallelism
Mixture of Experts: Sparse Computation for Efficient LLMs
RLHF and Preference Tuning: Aligning LLMs with Human Values
Small Language Models: When Less is More
Voice and Audio AI Models: Architecture, Training, and Deployment
Fine-Tuning LLMs: A Practical Guide to Tools, Techniques, and Best Practices
Vision Language Models: Architecture, Inference, and Practical Deployment
Inference & Deployment12
Test-Time Compute: Search, Verification, and Stopping
AI Safety Systems in Production: Policy, Permissions, and Validation
On-Device Multimodal AI: Text, Vision, and Audio Under a Memory Budget
Retrieval Systems Beyond Basic RAG: Hybrid Search, Reranking, and Evaluation
Modern Attention Kernels: Tiling, FlashAttention, and Paged Attention
Speculative Decoding: Draft, Verify, and Accept
Structured Generation: How LLMs Produce Valid JSON
Disaggregated LLM Serving: Separating Prefill from Decode
Inside the KV Cache: Memory Systems for LLM Inference
Edge AI Deployment: Running Models on Constrained Devices
NVIDIA TensorRT and Triton: Production LLM Inference
Advanced Techniques for Local AI Inferencing
Hardware & Systems7
Groq and Cerebras Accelerators: Deterministic Streaming and Wafer-Scale Compute
AI Cluster Networking: NVLink, Infinity Fabric, InfiniBand, and RoCE
The Diversification of AI Hardware: Beyond NVIDIA's Dominance
AMD ROCm for Machine Learning: The Alternative GPU Ecosystem
Apple's Metal Ecosystem for Machine Learning: Beyond MLX
Understanding GPUs and Parallel Computing
HPC Guide
Ecosystems & Tooling6
Model Context Protocol from First Principles: Messages, Capabilities, and Authorization
How torch.compile Turns PyTorch Programs into Fused Kernels
Apple MLX: A Complete Guide to Machine Learning on Apple Silicon
How Coding Agents Work: Inside the Agentic Loop
The Complete Hugging Face Ecosystem Guide
Modern Generative AI Guide