Applied AI Engineer · Researcher · Post Training
Mohammed Saqlain
PES University · Arctic Wolf
I work on post-training for small language models, RL and preference optimization across different agent harnesses and environments, with a focus on AI privacy and keeping models helpful.
About me
I am currently working at Arctic Wolf Networks, where I build agents that help with security operations. I graduated from PES University in 2026 with a B.Tech in Computer Science and Engineering (AI and ML).
On the research side I care about post-training for small language models, agent harnesses, and AI safety, especially privacy, contextual integrity, and keeping models helpful without brittle refusals.
Outside of that I play cricket, and I hold a Black Belt Dan 1 in Karate. I also sometimes write about philosophy, usually casually: the messy questions around knowledge, minds, and why some arguments feel solid while others slip.
News
- Oct 2026Paper on harness design for biomedical agents accepted to AgenticLS at NeurIPS 2026.
- Apr 2026HyperGNN work on neurodegenerative disorder classification accepted to ICTIS 2026.
- Jan 2026Joined Arctic Wolf Networks on the AI Security Operations team.
- 2025BiasNet published in Procedia Computer Science (FTNCT 2025).
- 2025Qualified for ACM ICPC Asia West Regionals.
Research interests
Post-training of small language models
I work on RL and preference optimization for small LMs, where controlled experiments are feasible and models often run close to personal data. The goal is training methods that improve behavior without relying only on prompting or brittle refusals.
Agent harnesses and RL environments
I study how the setup around a model (tools, retrieval, and episode design) changes what agents can learn and how they fail. I build environments so GRPO and related trainers can optimize agents on tasks with clear, checkable rewards.
AI privacy and contextual integrity
I am interested in teaching agents when information is appropriate to use or share, not only whether a citation or fact exists. I want rewards that penalize leaks while still rewarding helpful task completion, so models cannot score well by refusing everything.
Publications
View allReal PMIDs, Wrong Papers: Harness Design for Biomedical Claim-Verification Agents
Agentic AI for Biological Discovery (AgenticLS)
Mohammed Saqlain
HyperGNNs for Multi-Modal Classification and Severity Analysis of Neurodegenerative Disorders
K. Bhavish Raju, K. Musadiq Pasha, Mohammed Saqlain, Nishaan Padanthaya, Jayashree R
BiasNet: A Contrastive GNN Based Framework for Classifying Political Stance in News
T. Bhat, Mohammed Saqlain, A. Muralidharan, B. Das
Experience
I am currently working at Arctic Wolf Networks, with the AI Security Operations team, on agents for security operations. I have also worked at the NMCAD Lab at the Indian Institute of Science, building reinforcement-learning models for autonomous drone navigation and models for state-of-charge prediction in electric vertical take-off and landing aircraft.
Prior to that, I worked at ProjektAnalytics, where I built an AI compliance platform for medical use cases from the ground up. Earlier, at Boltzmann Labs, I worked on retrieval-augmented generation for healthcare.
Education
2022 – 2026
Bachelor of Technology, Computer Science and Engineering
Specialization: Artificial Intelligence and Machine Learning
Writing
View allAchievements
- ACM ICPC Asia West Regionalist (2025): Qualified for the ACM ICPC Asia West Regional Contest.
- Knight at LeetCode (May 2025): Achieved the title of Knight at LeetCode with a rating of 2000+.
- First Runner-Up, Flipr Hackathon 26.1 (Apr 2025): Secured the First Runner-Up position in the ML/AI track at Flipr Hackathon 26.1 for building an AI powered personal finance tracker.
- Black Belt Dan 1 in Karate (2019): Attained the rank of Black Belt Dan 1 in Karate, reflecting dedication, discipline, and expertise in martial arts.