A ML playground for education purpose

Fullstack machine learning inference template

A systems programming language where automatic differentiation is a compiler pass and model parameters are explicit, growable memory.

tpu-doc is a zero-dependency diagnostic binary for Google Cloud TPU environments that instantly validates hardware health, discovers software stack configurations, and provides AI-powered log analysis to eliminate expensive debugging downtime.

A Branchless, Zero-Jitter Ingress Router for 32-GPU Distributed Mesh Networks utilizing JAX/XLA and NCCL.

Compiler MVP that detects Transformer fusion patterns, generates optimized CUDA kernels with WMMA Tensor Cores, and executes them on real GPU hardware — 10.5 TFLOPs on RTX 2070, correctness validated against PyTorch.

A custom Deep Learning Framework built from scratch using NumPy. Implements Automatic Differentiation, Backpropagation, and modular Layer architectures.

Analytical benchmark connecting pipeline parallelism and KV cache management: bubble-time overlap for disaggregation, prefix reuse tax of PP stages, and tiering speedup across NVLink, InfiniBand, and PCIe

FlashAttention v1 forward pass in CUDA for NVIDIA Turing (SM75)

Discrete-event simulation benchmark for graceful degradation strategies in LLM serving: output capping, CoT disabling, model downgrade, and selective drop under progressive pressure.

🔍 Explore machine learning hands-on with scikit-learn through projects that illustrate key concepts like classification, regression, and clustering.

Distributed, GPU-aware workload scheduler for heterogeneous clusters: queueing, quotas, GPU flavors, and autoscaling.

Analytical benchmark for sliding window attention KV cache management: quality vs window size tradeoffs, SWA vs eviction comparison, prefix sharing interaction, and operational window recommendations across four attention distributions

Tungsten Alpha: A JAX-native, 7-layer Adelic-Riemannian Operating System. Implements resurgent gradient flow through discrete logic and Levi-Civita parallel transport for decentralized XLA clusters. Hardened for zero-jitter, systolic array execution.

Simulation benchmark for canary deployment strategies in LLM model updates, measuring safety, rollback behavior, bad exposure, and rollout speed across good, bad, and marginal candidate scenarios

A research-grade course in Numerical Methods and Optimization for ML. No deep learning libraries—just pure math and NumPy. Rebuilds automatic differentiation engines, stable SVD/Cholesky algorithms, quasi-Newton solvers (BFGS), stable log-sum-exp layers, and Gaussian Process regressions from scratch

Discrete-event simulation for LLM serving capacity planning: how many GPUs for a p99 TTFT SLO under ShareGPT traffic, batching fragmentation, and autoscaling lag — with cliff analysis and three levels of planning rigor.

Monte Carlo simulation benchmark for MoE expert activation memory variability and admission control under non-deterministic routing — including routing skew amplification, top_k sweep, and policy comparison across uniform, skewed, and very_skewed distributions

Discrete-event simulation benchmark for LLM request cancellation handling, measuring wasted compute, KV residency overhead, and survivor latency recovery across polling and push-based strategies.

Discrete-event simulation benchmark for decode preemption policies in LLM serving: when to interrupt a running decode to admit an urgent prefill, and whether checkpoint or recompute is cheaper.