FAHIM FAISAL

Conducted PhD research on multilingual training and evaluation across BERT-style encoders to large dense and MoE LLMs, utilizing Hugging Face Transformers and parameter-efficient fine-tuning to inform next-generation model design.

Designed reproducible evaluation pipelines for 100+ language varieties, covering reasoning, toxicity, QA, and machine translation, built with SQL-backed data stores and Grafana visualization, ensuring consistent benchmark reporting across experiments.

Deployed and served LLMs on GMU GPU clusters and AWS SageMaker, Microsoft Azure, and Google Cloud Platform, automating scaling with Docker, Kubernetes, and CI/CD pipelines, reducing inference latency and operational costs.

Performed statistical analysis of model behavior using Python and scientific ML libraries like SciPy, conducting calibration and significance testing in ambiguous research contexts to identify key performance gaps and guide subsequent model refinements.

Developed agentic AI pipelines for multilingual reasoning alignment through expert distillation and reinforcement-learning post-training using PPO in PyTorch and TensorFlow on GCP, documenting results in an internship paper.

Evaluated LLM benchmarks with lm-evaluation-harness across 100+ language varieties, performing calibration and robustness analysis to enhance model reliability.

Built Docker containers on AWS SageMaker to ensure reproducible evaluation and inference for diverse LLM benchmarks and research experiments.

Implemented the Pinterest Pixie algorithm for aspect-based recommendation on large-scale eBay user-item-aspect bipartite graphs, training on 1.3M interactions across three product categories.

Achieved improved offline ranking metrics (NDCG, MRR, Recall@K) and designed Spark-SQL data pipelines for efficient data processing.

Led safety and policy alignment for an in-house e-commerce LLM, generating 1M+ synthetic preference examples using Qwen, Phi, LLaMA, and GPT.

Built an automated compliance evaluation that identified refusal-quality issues and reduced false-positive tradeoffs, enhancing LLM reliability and ethical deployment.

Contributed to community-engaged project design, focusing on the social impact of technology and research initiatives.

Participated in a project investigating the perspectives of parents of students with disabilities during the COVID-19 pandemic.

Assisted professors with course material development, grading assignments, and facilitating lab sessions for computer science students.

Provided individualized support and mentorship to students, clarifying complex concepts and fostering a deeper understanding of technical subjects.

Managed administrative tasks related to course delivery, ensuring smooth operation and effective communication with students and faculty.

Assisted faculty in conducting research, including data collection, analysis, and literature reviews in computer science domains.

Supported undergraduate courses by providing teaching assistance, leading tutorial sessions, and guiding students on academic projects.

Contributed to research projects, assisting with experimental setup, data processing, and result interpretation.

Provided academic support to students, offering guidance on assignments and clarifying complex technical concepts.

Developed and implemented software components, contributing to the lifecycle of R&D projects.

Collaborated with team members to design, test, and deploy software features, ensuring adherence to quality standards.