Skip to content
View NoviceCoderInfinity's full-sized avatar

Highlights

  • Pro

Block or report NoviceCoderInfinity

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse

Anupam Rawat — ML Systems & Applied AI

IIT Bombay EE (B.Tech + M.Tech) · Dual Minor CSE + CMInDS · IBM Research (neural operators / deep RL)

I build research-grade PyTorch systems and ship production AI — from mesh-agnostic neural operators to structure-aware LLM inference and real-time multimodal agents.

🌐 Portfolio · LinkedIn · Email


Now

Track Focus
IBM Research COSMOS — MoE neural operator (NeurIPS 2026 workshop target) · ARGO — deep RL orchestrator for hybrid PDE solving (ICLR 2027 target)
PriorityKV Structure-protected mixed-precision KV cache for long agent traces (FlashInfer, H200)
Product AI Ghar (voice multimodal home design) · ReVival (circular commerce agents)

Featured projects

PriorityKV — ML systems / inference

Structure-aware BF16/INT4 KV compression so tool schemas and instruction hierarchies survive long multi-turn agent traces. Structure keep 0.688 vs FixedHot 0.125 at matched budget; FlashInfer LSE/packed parity on NVIDIA H200.

Ghar — real-time multimodal

Voice AI interior designer on Gemini Live API: describe a home → concepts → narrated room tours → architect Build Pack. Short demo · Full demo

ReVival — Amazon HackOn S6

7-agent circular-commerce pipeline (vision grading, disposition, pricing, buyer matching) on FastAPI + AWS Bedrock. Live demo

IBM Research (ongoing)

  • COSMOS — discretization-agnostic MoE neural operator; compositional PDE benchmark; reconstruction nMSE cut 180×; beats FNO / U-Net / persistence on 20-step rollout (Wilcoxon (p \approx 10^{-46}))
  • ARGO — PPO-from-scratch RL system composing neural operators + classical solvers; 4×A100 DDP (10M steps / 6.9h)

Stack

PyTorch · Deep RL (PPO/GAE) · Neural Operators · FlashInfer · Transformers / LLMs · DDP / torchrun · HPC/LSF · FastAPI · AWS · Gemini Live API


Selected results

  • Structure-aware KV keep 0.688 vs FixedHot 0.125 (PriorityKV, H200)
  • COSMOS: beats FNO / U-Net / persistence on 20-step rollout (Wilcoxon (p \approx 10^{-46}))
  • Wyzr Fintech: +15% categorisation accuracy, +33% reconciliation match rate (production LLM)
  • GeoNLI (ISRO / Inter IIT): 82% captioning · 83.3% VQA · 64.9% grounding

Selected competitions

  • Amazon HackOn Season 6 — ReVival (circular commerce)
  • Qualcomm Innovation Challenge — Reflection Removal (Rank 4)
  • Kaggle Image Super Resolution — Rank 4
  • Inter IIT Tech Meet 14.0 — ISRO Earth-Observation (GeoNLI)

Pinned Loading

  1. satushparikh/CS726-assignments satushparikh/CS726-assignments Public

    Jupyter Notebook 1

  2. CS_754_Advanced_Image_Processing CS_754_Advanced_Image_Processing Public

    TeX

  3. shoryasethia/RobustSIRR shoryasethia/RobustSIRR Public

    Qualcomm VisionX, Team Name : ClarifyAI, Rank : 4

    Python 2