Software Engineer working on the LLM team at Anyscale!

Current Project: SkyRL; Skythought

• One of the core contributors of SkyRL: https://github.com/NovaSky-AI/SkyRL , building a full stack library for post-training LLMs

• Core contributor to https://novasky-ai.notion.site/skyrl-v0 - implemented a scalable remote server for RL training on SWE-Gym, contributed to building asynchronous multi-turn rollout implementation to improve SWE-Bench performance of Qwen-3-8B by 5.8%.

• Core contributor to novasky-ai.notion.site/skyrl-sql, one of the first open-source models trained with multi-turn RL on Text2SQL - matching GPT-4o and o4-mini on the Spider benchmark.

• One of the core maintainers for the Skythought repo: https://github.com/NovaSky-AI/SkyThought/commits?author=SumanthRH

• Worked on standardized, scalable evaluation for reasoning models

Past Project: LLMForge, Anyscale's fine-tuning framework

• Added support for different fine-tuning tasks (such as instruction tuning and causal LM)

• Improved model source support to allow bringing any HuggingFace model with any chat template to fine-tune on Anyscale.

• Preference tuning, Function calling fine-tuning

• Improved DPO training speed by 20-40% with prefix sharing: https://github.com/frankxwang/dpo-prefix-sharing

• Led building an SDK for models trained on Anyscale: https://docs.anyscale.com/reference/llm_models

Working on language models in Prof. McAuley's Lab at UCSD

• Teaching Assistant for CSE 232: Principles of Database Systems and CSE 21: Mathematics for Algorithms and Systems.

• Responsibilities included conducting weekly discussion sessions, preparing question papers for examinations, etc. Best part: Office hours!

• Set up a finetuning codebase for language models from scratch for use in C3’s Generative Search application

• Features: Support for difference causal and sequence-2-sequence models, ability to mix different training datasets (for a text-to-text or a causal language modelling task), visualize metrics on multiple evaluation datasets, parameter-efficient fine-tuning and quantization support, etc

• Trained 10B+ parameter models on 1M+ samples using DeepSpeed and 🤗 Accelerate.

Worked on video-based object detection methods to improve Hakimo's Remote Guarding solution