Agentic Ops · Model Evaluation · RLVR Datasets
Technical Manager, AppliedAI at Pareto
I build RLVR training datasets and RL environments that frontier AI labs use to train and evaluate their models, grounded in eight years of research in data mining and LLM evaluation.
Fahim Arefin works at the intersection of frontier-model training and computer science research. At Pareto, he devises the strategy and groundwork for off-the-shelf RLVR training datasets and RL environments for frontier labs, and builds agentic pipelines that automate delivery operations end to end: intake, QA, verification, and shipping.
Before Pareto, he spent three years at Turing leading delivery programs for LLM training-data engagements with frontier AI labs across code, reasoning, and STEM domains. In parallel, he serves as Lecturer in Computer Science and Engineering at the University of Dhaka, where he teaches, advises students, and supervises research on data mining, LLM capabilities, and AI in education.
His research spans sequential pattern mining, LLM evaluation for competitive programming, and applied machine learning, published in venues including ACM TOSEM, Expert Systems with Applications, PAKDD, and ICSE workshops. He co-founded AlterYouth, a C2C scholarship platform funding 2,000+ monthly scholarships for underprivileged students in Bangladesh.
Pareto.AI
Devising strategy and laying the groundwork for off-the-shelf RLVR training datasets and RL environments for frontier labs. Building agentic pipelines that automate delivery ops end to end: intake, QA, verification, and shipping.
Read: Verifier engineering is the moat ↗University of Dhaka
Teaching CS courses and supervising research on data mining, LLM capabilities, and AI in education. Student Advisor for the department since Nov 2024. Previously Part-time Lecturer (Jul 2022 – Mar 2023).
Turing.com
Held full P&L accountability for concurrent LLM training-data programs with frontier AI labs, maintaining 100% client retention while expanding gross margin. Scaled campaigns from 0 to 500+ trainers generating $4M in monthly revenue across code, reasoning, and STEM domains. Designed eval datasets that exposed model failure modes, with outcomes guiding clients' successive fine-tuning cycles.
Read: Closing note on my Turing chapter ↗AlterYouth
C2C scholarship platform enabling anyone in the world to fund the education of underprivileged students in Bangladesh through digital banking. Funds 2,000+ scholarships every month, disbursing 35M+ BDT in scholarships annually.
Start a scholarship ↗Beetles Cyber Security
Developed CrowdSpark, a proprietary engagement-management application to control, organize, and visualize the pentesting process.
CrowdSpark ↗ACM Transactions on Software Engineering and Methodology (TOSEM)
Tabassum, Hossain, Arefin, Islam, Zaman
Data in Brief (Elsevier)
Lina, Mahmud, Hasan, Arefin, Rizvee, Rahman, Khan
IEEE/ACM LLM4Code @ ICSE 2025
Hossain, Tabassum, Arefin, Zaman
Expert Systems with Applications
Rizvee, Ahmed, Arefin, Leung
IMCOM 2024
Solaiman, Rahman, Arefin, Ahmed, Leung, Madill
IMCOM 2024
Abid, Shaha, Kabir, Arefin
Dhaka University Journal of Applied Science and Engineering
Arefin, Islam, Ahmed
IMCOM 2022
Arefin, Ahmed, Rizvee, Leung, Cao
IEEE UEMCON 2022
Rizvee, Arefin, Abid
IEEE R10 HTC 2022
Jabin, Suhi, Arefin, Hasib
IEEE IEMCON 2022
Rizvee, Arefin, Khan, Islam, Rabbi
STI 2021
Islam, Hossain, Arefin
IEEE IEMCON 2021
Sakib, Hasib, Tasawar, Tanzeem, Arefin, Islam
RLVR dataset strategy, RL environment design, verifier engineering, rubric and QA systems, expert pipeline scaling for code, reasoning, and STEM domains.
Agentic pipeline architecture for delivery operations: automated intake, evaluation, verification, and shipping workflows.
Sequential pattern mining, LLM evaluation methodology, sentiment analysis, applied machine learning; 15 peer-reviewed publications.
Program delivery for frontier-lab engagements, cross-functional team management, university teaching, research supervision, student advising.
Interested in RL environments, training data partnerships, research collaboration, or teaching? Always happy to connect.