System
Production RAG and model adaptation
Delivered legal-tech RAG and QLoRA fine-tuning work before focusing fully on LLM systems.
LLM systems · CUDA kernels · Inference
LLM inference, made measurably faster.
I work where the model meets the metal: attention kernels, speculative decoding, KV-cache behavior, compression, and the serving stack around them. Every claim on this site links to code, profiling traces, and benchmarks.
- Attention throughput
- 0.0×
- Decoding speedup
- 0.0×
- Memory reduction
- 0%
- LLM from scratch
- 0M
Attention throughput
vs PyTorch SDPA, A100
Decoding speedup
Identical output dist.
Memory reduction
Tensor-network MPO
LLM from scratch
Hybrid Mamba-2 + attn
fig. 01 — tiled causal attention sweep
liveBr=64 · Bc=64 · SRAM-resident · HBM traffic O(N²) → O(N)read the kernel write-up →
45-second signals
Six things worth your scroll
Timeline
Milestones
Research, systems, benchmarks, and open-source across the lab record.
2023
2024
System
Project Chimera begins
Designed a 700M hybrid Mamba-2 and Transformer LLM, tokenizer, training schedule, and evaluation path.
System
VAANI offline assistant
Built a Hindi-first offline voice stack with wake word, ASR, local LLM, and TTS under 800ms.
2025
Benchmark
Speculative decoding reaches 2.4x
Implemented draft-verifier inference with adaptive lookahead and distribution-preserving sampling.
Open Source
vLLM disaggregated prefill patch
Fixed request-ID mismatch across prefill and decode nodes causing KV cache transfer hangs.
Benchmark
Sparse MoE scaling run
Implemented top-k routing, z-loss balancing, entropy logging, and matched-FLOP dense baselines.
2026
Research
Grokking Beyond Addition published
Circuit-level study across abelian and non-abelian algebraic operations published on Zenodo.
Benchmark
FlashAttention-2 kernel profiled
CUDA C++ tiled attention kernel reached 2.1x throughput over PyTorch SDPA at sequence length 4096.
Research
CompactifAI extension completed
Extended tensor-network compression with adaptive bond-dimension scheduling and model healing.
Benchmarks
Measured systems signals
Speculative decoding2.4x
FlashAttention-2 kernel2.1x
Sparse MoE2.3x
API p95 reduction35%
MPO memory reduction93%
Systems map
Training, compression, kernels, serving, evaluation
Current investigations
Internal notebook surface
Failed Experimentsfailed
Non-Abelian Grokking Capacity Ceiling
Can the same one-layer transformer that groks abelian operations grok non-abelian groups under longer training?
Compression Studiesreproduced
Uniform MPO Compression Collapse
Does a single global bond dimension preserve quality across all transformer layers?
Benchmark Logsarchived
Fixed-Gamma Speculative Decoding
Is a fixed draft lookahead window enough for CPU speculative decoding?
Scaling Studiesreproduced
Sparse MoE Router Entropy
Can routing entropy predict expert collapse before validation loss exposes it?
Contract engagements
Serving models in production and the GPU bill hurts?
Fixed-scope inference optimization sprints with a benchmark target agreed up front — and a guarantee behind it.
See engagements