Hi, I am Jebish.

NLP Researcher & Machine Learning Engineer

I am passionate about multilingual NLP, Human-centered LLMs/VLMs, and agentic AI systems. I am currently leading research at Cohere Labs Community and M2ai, working on large-scale language model evaluations and low-resource language processing.

My current aim is to study how language models generalize, reason, and adapt across contexts. At the same time, I also develop workflows and agents based systems dependent on these models at my workplace - Dogma International.

About

I am an AI researcher and machine learning engineer with expertise in multilingual NLP, vision-language models, and agentic AI systems. I hold a Bachelor of Engineering in Electronics and Communication (Valedictorian) from Tribhuvan University, Nepal.

Currently, I lead community research at Cohere Labs and spearhead multilingual NLP initiatives at M2ai, with fundings obtained from OpenAI, Anthropic, and Cohere. My research spans cultural representation in VLMs, multilingual benchmarking, regulatory NLP, and agentic AI frameworks.

Research Experience

ML Agents Community Lead
Cohere Labs · June 2025 – Present

Leading 20+ researchers on large-scale benchmarking of LLM reasoning across 50 task categories. Designed evaluation pipelines for 6 reasoning approaches (CoT, SoT, BoT, LtM, CoVE, GoT) across 8 language models.

Researcher
M2ai · Sep 2024 – Present

Leading research on multilingual, multimodal, and low-resource NLP with $20,000+ grants from OpenAI, Anthropic, and Cohere. Developed Mantra-14B, a state-of-the-art Hindi-English bilingual model. Built AI-generated text detection across 23 languages and 12 generators.

Research Fellow
Traversaal.ai · Dec 2024 – May 2025

Architected AgentPro, a REACT-based agentic framework for complex data science workflows. Constructed comprehensive benchmarks across 8 task categories to evaluate agentic architecture performance.

Community Researcher
Cohere Labs · Sep 2023 – May 2025

Led study on cultural knowledge disparities in VLMs across 200+ countries. Co-developed INCLUDE, multilingual benchmark for localized knowledge. Contributed to Kaleidoscope multimodal evaluation benchmark.

Industry Experience

Associate AI Engineer
Dogma International · June 2025 – Present

Co-developed Scottish Charity Agent with RAG-based pipeline for compliance reviews. Deployed Feasibility Analysis Agent on Azure. Engineered OCR and LLM pipeline for penalty charge extraction.

Teaching

Instructor
National College of Engineering & Pulchowk Campus · Jan 2024 – Aug 2025

Instructed undergraduate students in Artificial Intelligence, Embedded Systems, Numerical Methods, Computer Programming, and Digital Signal Processing. Managed curriculum delivery and lab sessions.

Publications

INCLUDE: Evaluating Multilingual Language Understanding with Regional Knowledge
ICLR Spotlight
Angelika Romanou, Negar Foroutan, ..., Jebish Purbey, ..., Antoine Bosselut
Uncovering Cultural Representation Disparities in Vision-Language Models
IJCNLP-AACL 2025 Findings
Jebish Purbey, Ram Mohan Rao Kadiyala*, Siddhant Gupta*, et al.
Evaluating Generalization Capabilities of LLM-Based Agents in Mixed-Motive Scenarios
NeurIPS
Chandler Smith, Marwa Abdulhai, ..., Jebish Purbey, ..., Ziyan Wang
DSBC: Data Science task Benchmarking with Context Engineering
IJCNLP-AACL 2025
Ram Mohan Rao Kadiyala, Jebish Purbey, Siddhant Gupta, et al.
Lexical Reranking of Semantic Retrieval (LeSeR) in Regulatory Domains
RegNLP COLING
Jebish Purbey, Drishti Sharma, Khawaja Murrad, et al.
Global PIQA: Evaluating Physical Commonsense Reasoning Across 100+ Languages
Preprint
Tyler A. Chang, Catherine Arnett, ..., Jebish Purbey, ..., David Ifeoluwa Adelani

Skills

Large Language Models
Natural Language Processing
Computer Vision
Multilingual AI
RAG Systems
Agentic AI
Fine-tuning
Benchmarking
Python
PyTorch
Transformers
Azure

Get In Touch

Email: jebishpurbey@gmail.com

Location: Kathmandu, Nepal

Professional Links

LinkedIn: linkedin.com/in/jebishpurbey

Google Scholar: Google Scholar Profile

Github: github.com/jebish