Mihir Sahu profile photo 1 on Tal

Mihir Sahu

Senior SWE @Arcana | ex-Graviton (HFT) | AI/ML Inference | Distributed Systems | IIT Patna Silver Medalist

Bengaluru

2+ years of work experience

Arcana
Jul 20262 mosPresent

Bengaluru, Karnataka, India

Graviton Research Capital LLP
Nov 20258 mosJun 2026

HFT
UnifyApps
Jun 20241 yr 5 mosOct 2025

Gurugram, Haryana, India

Founding ML Engineer and employee ~50. Primarily worked on orchestration and harnesses for AI Agents. Built agents for enterprises bringing in 1.5M+ $ ARR. Also worked on a range of other ML stuff: - Led the development of Graph based RAG (GraphRAG). Owned the system end to end from indexing to retrieval - GraphRAG became a critical component of the Enterprise Context Graph and Personalization Graph used for Enterprise Search. - Distributed training and inference: LLMs and Vision models - Inference optimization: Distillation, Quantization, etc - AI Agents: Long term memory, Multi Agent Systems with parallel task execution, speculative tool calling, context prefetching, etc - Large scale knowledge graphs and clustering with de-duplication. - Classical ML/NLP/Computer Vision - MLOps, ML platform/infra.
UCF Center for Research in Computer Vision (CRCV)
Jan 20246 mosJun 2024

Orlando, Florida, United States

Worked on reasoning capabilities of Large Vision Language Models on compositional questions.
Razorpay
May 20233 mosJul 2023

Bangalore Urban, Karnataka, India

- Designed and implemented a Retrieval Augmented Generation (RAG) System equipped with a semantic cache for latency optimization. - Integrated data pipelines to enable continual learning. - Implemented Server-Sent Events to reduce wait times by over 75%. - Awarded PPO based on performance.
GitHub
@Mihirsahu2307

Mihir’s AI Stack

antigravity
antigravity
codex
codex
cursor
cursor

Education

Indian Institute of Technology, Patna

Bachelor of Technology - BTech, Electrical and Electronics Engineering

Indian Institute of Technology, Patna

20204 yrs2024

Join Mihir and 10k+ other professionals on

Tal Boss

Tal is the app where bosses meet engineers & builders directly, without middlemen

Download Tal Boss

Mihir Sahu on Tal

Headline

Senior SWE @Arcana | ex-Graviton (HFT) | AI/ML Inference | Distributed Systems | IIT Patna Silver Medalist

About

Broad interests: Startups, Distributed Systems, Optimization, Machine Learning, Quantitative Finance Obsolete: I currently work as a Software Engineer at Graviton, a leading HFT firm. Prior to this, I had joined UnifyApps at seed stage while it was still a 50-ish person team. As an early AI/ML Engineer, I worked on the core ML infrastructure and orchestration harness for AI agents. During my undergrad at IIT Patna, I have contributed to ML research, focusing on areas such as Representation Learning and Multimodal Learning, with first-author publications in prestigious international conferences. My research and project experience extend to Information Retrieval, Ranking Systems, and Large Multimodal Models. An avid problem-solver, I have excelled in competitive programming contests such as ICPC, Meta Hacker Cup, and Google Kick Start. I take a keen interest in exploring topics related to Probability, Randomness, and Contemporary Philosophy.

Location

Bengaluru

AI stack

antigravity

codex

cursor

Work experience

Senior Software Engineer at Arcana, Jul 2026 - Present · 2 mos, Bengaluru, Karnataka, India

Software Engineer at Graviton Research Capital LLP, Nov 2025 - Jun 2026 · 8 mos, HFT

Product Engineer (AI/ML) at UnifyApps, Jun 2024 - Oct 2025 · 1 yr 5 mos, Gurugram, Haryana, India, Founding ML Engineer and employee ~50. Primarily worked on orchestration and harnesses for AI Agents. Built agents for enterprises bringing in 1.5M+ $ ARR. Also worked on a range of other ML stuff: - Led the development of Graph based RAG (GraphRAG). Owned the system end to end from indexing to retrieval - GraphRAG became a critical component of the Enterprise Context Graph and Personalization Graph used for Enterprise Search. - Distributed training and inference: LLMs and Vision models - Inference optimization: Distillation, Quantization, etc - AI Agents: Long term memory, Multi Agent Systems with parallel task execution, speculative tool calling, context prefetching, etc - Large scale knowledge graphs and clustering with de-duplication. - Classical ML/NLP/Computer Vision - MLOps, ML platform/infra.

AI Researcher at UCF Center for Research in Computer Vision (CRCV), Jan 2024 - Jun 2024 · 6 mos, Orlando, Florida, United States, Worked on reasoning capabilities of Large Vision Language Models on compositional questions.

Software Engineer at Razorpay, May 2023 - Jul 2023 · 3 mos, Bangalore Urban, Karnataka, India, - Designed and implemented a Retrieval Augmented Generation (RAG) System equipped with a semantic cache for latency optimization. - Integrated data pipelines to enable continual learning. - Implemented Server-Sent Events to reduce wait times by over 75%. - Awarded PPO based on performance.

Education

Bachelor of Technology - BTech, Electrical and Electronics Engineering, Indian Institute of Technology, Patna

Public links

github: https://github.com/Mihirsahu2307