Research interests: Currently in AI Safety
Also Working on simplifying fine-tuning jobs for researchers with TRLoom.
- Real PMIDs, Wrong Papers: Harness Design for Biomedical Claim-Verification Agents has been accepted to the Agentic AI for Biological Discovery workshop (AgenticLS) at NeurIPS 2026.
TRLoom — YAML-driven end-to-end fine-tuning on Hugging Face TRL.
pip install trloom
trloom run config.yaml # local
trloom run config.yaml --modal # remote GPUsDocs: saqlain2204.github.io/trloom · PyPI: trloom If this saves you setup time, a ⭐ helps others find it.
Currently at Arctic Wolf Networks.
Previously:
- Built RL models for autonomous drone navigation (PPO) and predictive battery State-of-Charge models as a Research Intern at IISc
- Designed RESTful APIs and a hybrid-retrieval RAG chatbot as an AI Engineer Intern at ProjektAnalytics
- Built a RAG pipeline for clinical query answering and a live speech-to-text application as a Research Intern at Boltzmann Labs
- Contribute to FOSS across the AI ecosystem, including Hugging Face Diffusers/Transformers
- Shipped projects spanning agentic CLI assistants, hate-speech classification with explainability, LLM-based topic modelling, and automated content pipelines
- Real PMIDs, Wrong Papers: Harness Design for Biomedical Claim-Verification Agents — AgenticLS @ NeurIPS 2026
- HyperGNNs for Multi-Modal Classification and Severity Analysis of Neurodegenerative Disorders — ICTIS 2026
- BiasNet: A Contrastive GNN Based Framework for Classifying Political Stance in News — FTNCT 2025
- Reading Between the Lines: LLM-Powered Topic Modelling and Graph-Based Insights from Research Abstracts — WCAIAA 2025
- Comparative Analysis of Traffic Accident Detection with Emphasis on Explainability of DL Models — ICMBDC 2024
- ACM ICPC Asia West Regionalist - 2025
- Flipr Hackathon 26.1 - Runners Up
- LeetCode Guardian - @activation-tanh
Open to conversations on pre-trianing and post-training of language models — reach out via email or LinkedIn.