3D Perception
LiDAR, point clouds, SLAM, odometry, reconstruction & Gaussian splatting.
Head of Engineering · Inkers · Bengaluru
I build deep-learning systems, 3D perception, and high-performance infrastructure — custom neural networks, CUDA kernels LiDAR / point-cloud pipelines that reconstruct the physical world and ship to production.
LiDAR, point clouds, SLAM, odometry, reconstruction & Gaussian splatting.
Custom neural networks, computer vision, GenAI & structural-defect detection.
CUDA kernels, kernel fusion, quantization & C++/gRPC services.
EMLO V2–V4 (MLOps) & EVA 4.0 (CV) at The School of AI. 287 repos.
AI market-research platform for Indian equities — a unified 14-point scoring framework fusing technical, fundamental, news-sentiment & derivatives data across 850+ NSE stocks.
A mobile-first AI nutrition companion — scan packaged-food barcodes for instant ingredient analysis, additive safety warnings & macro breakdowns, built for Indian consumers.
AI platform for construction & project management — LiDAR reconstruction, 3D defect tracking, and an in-browser point-cloud viewer.
A full-blown project & construction-management copilot — GenAI agents reasoning over construction data.
Ask my AI
I built an AI trained on my work, writing, and résumé — it knows my projects, experience, and how I think. Ask it anything about me, in my voice.
ai.satyajit.siI write in-depth explainers of AI papers, models and systems on ai.satyajit.si (701 articles so far). The latest:
Alibaba AMAP's ABot-Recon (arXiv 2608.27529) reconstructs camera paths and point maps from long video while attending to only the current frame and the 11 before it.
Meta's Astryx is a React 19 and StyleX design system, MIT-licensed, in beta, pitched as built for people and the agents working with them. It launched in June; two posts this week recirculated it.
vLLM's Kimi-K3.1-Pro-pruned75 keeps 224 of the 896 routed experts in each of Kimi K3.1 Pro's 92 MoE layers, chosen without data.
The TSAI Learning Platform
Building LLM agents with LangChain and LlamaIndex, from context-aware chains and tools to serving them as a FastAPI service on AWS.
A tour of multimodal LLMs (GPT-4V, Flamingo, LLaVA) and serving BakLLaVA behind an OpenAI-compatible FastAPI backend on AWS.
Leading engineering for industrial-AI products: 3D perception, LiDAR/point-cloud processing, and structural-defect analysis shipped as C++/CUDA/gRPC services.
3D mesh processing, reconstruction, and deep-learning pipelines for construction-site monitoring and defect detection.
Built the Observance.AI full-stack construction-AI platform and the 3D reconstruction pipeline from the ground up.
Edge computer-vision: real-time stereo processing and model optimization on NVIDIA Jetson.
Designed and taught EMLO V2, V3 & V4 — end-to-end production MLOps — and contribute to the EVA 4.0 computer-vision curriculum.