TUSHAR VERMA

Applied Researcher at eBay.
Building and deploying image, text, and multimodal models.

I work with image models, BERT-style text models, and vLLM deployments for text and multimodal LLMs. Previously at Convin, I trained and deployed automatic speech recognition systems.

NOW
EBAY / DEC 2025 - PRESENT
PREVIOUSLY
CONVIN / FEB 2022 - NOV 2025
INTERESTS
ML / DL / PYTORCH / TRITON / VLLM

01 / RESEARCH & PAPER(S?)

Abstract blue network visualization

MULTILINGUAL ASR

A speech recognition system for low-resource, noisy, code-mixed Indian languages. The approach fine-tunes pretrained models with text transliterated to Devanagari and groups similar-sounding characters.

AREA
SPEECH / NLP
STATUS
PUBLISHED

02 / WRITING

RECENT ARTICLES

03 / NOTES

WORKING NOTES

Short notes on machine learning, deep learning, PyTorch, Triton, vLLM, and experiments will live here.