REMOTE#01-CORE

Abhishek Ghosh

GPU Systems Engineer • Triton Kernels • Full-Stack Design Technologist

Building high-throughput GPU computing pipelines, custom OpenAI Triton flash-attention kernels, and low-latency serving backends. Obsessed with mechanical precision, hardware SRAM tiling, and tactile matte-white digital artifacts.

Projects

github.com/codenuclei

GPU REINFORCEMENT•2026

Kernel Crucible: Autonomous Silicon Evolution

Autonomous silicon evolution coliseum with cyberpunk frontend and RLVR GRPO reinforcement learning engine for auto-generating, profiling, and benchmarking high-throughput GPU kernels.

GPU ACCELERATION•2026

Optimized GPU Math Kernels for ML Inference

Handcrafted SRAM-tiled forward & backward flash-attention operators and matrix multiplication kernels in Triton and CUDA. Zero register spilling with continuous batching integration.

SIMULATION & PROFILER•2026

CUDA Visualizer & Accel-Sim GPU Profiler

Interactive browser playground and architectural simulation profiler for writing CUDA kernels and inspecting warp divergence, memory coalescing, and cache hits without physical GPU hardware.

VECTOR RETRIEVAL•2026

Multi-Modal Video & Image Search Indexer

Multi-modal drive vector indexing and search engine covering images, video keyframes, and PDFs. Features face recognition clustering and high-precision semantic visual queries.

Internships

VERIFIED RECORD

AI RESEARCH•2024 – 2025

SDE Intern at Google

AI Research (NDA)

Engineered machine learning systems and deep learning research pipelines under non-disclosure governance. Benchmarked model latency, optimized tensor transformations, and received the Outstanding Intern Award.

FULL-STACK & PLATFORM•JUNE – SEPTEMBER 2026

Software Development Intern

Masters' Union • DLF CyberPark, Gurgaon

Worked across AI-powered internal products using TypeScript, Next.js, Python, and modern ML tooling, including Drive FaceIndexer for intelligent face-based media organization and search.

Open Source

CONTRIBUTIONS

CORE CONTRIBUTOR•KORNIA-RS

Kornia: High-Performance GPU Computer Vision

kornia / kornia-rs • Rust, CUDA & SIMD

Built a native CUDA FAST corner detector in Rust, achieving 2.8×–3.5× CPU speedups with allocation-free execution. Reworked AVX2/NEON SIMD kernels for bit-identical parity and 31.8% higher throughput.