Qasim Rehman

Dissertation Internship at BioAscend

Designed and implemented an end-to-end ML pipeline for QSAR prediction under extreme data scarcity; trained graph neural networks (GIN, RGCN, GCNP) with scaffold-based data splitting to ensure robust evaluation and mitigate data leakage; applied self-supervised pretraining to improve generalization to unseen molecular scaffolds; developed reinforcement learning workflows for goal-directed molecular generation with QSAR as reward; designed multi-objective optimization frameworks incorporating domain and business constraints (drug-likeness, synthetic accessibility, chemical validity).

Software Engineer at Self Employed

Designed and deployed data-driven systems for manufacturing clients, supporting operational workflows; developed forecasting and predictive analytics solutions to support production planning and resource optimization; collaborated with domain stakeholders to translate operational constraints into analytical models and system requirements.

Software Engineer at InvoZone

Developed and maintained backend services for data-driven applications, focusing on scalable APIs and reliable data handling; integrated backend systems with frontend components to support responsive, user-facing applications in a production environment; collaborated within cross-functional teams to design and deploy features involving real-time data processing and system integration.

Intern, Dissertation at Bio Ascend

Designed and implemented end-to-end ML pipeline for QSAR prediction under extreme data scarcity, applying ML engineering best practices for training, validation, and generalization. Trained graph neural networks (GIN, RGCN, GCN) with scaffold-based datasets to ensure robust evaluation and mitigate data leakage. Applied self-supervised pretraining to improve generalization to unseen molecular scaffolds with limited labelled data. Developed reinforcement learning workflows for goal-directed molecular generation, integrating QSAR as reward signals. Designed multi-objective optimization frameworks incorporating domain and business constraints (drug-likeness, synthetic accessibility, chemical validity). Communicated findings, risks, and trade-offs to academic and industry stakeholders to support informed decision-making.