• Designing offline scientific-ml benchmarks for pde solver discovery and stable-structure search, with protected budgets, hidden transfer splits, and reproducible run evidence.
• Instrumenting long-horizon agent runs across trajectories, tool calls, checkpoints, verifier outcomes, cost, and failure stages to tell real benchmark progress from lucky runs.
• Building evaluation infrastructure that separates invalid submissions from genuine zero scores and pairs diagnostic probes with hidden authoritative transfer evaluation.

• Investigating whether naturalistic strategies for eliciting and resisting llm sycophancy map to distinct internal mechanisms across llama- and mistral-family models.
• Building a controlled 1,500+ prompt benchmark and reproducible pipeline for activation probing, cross-strategy transfer, and representation-level analysis.
• Testing causal steering interventions with preservation checks for factual accuracy, instruction following, and unintended capability degradation.

• Owned implementation of a client-facing tender-to-opportunity pipeline that extracted government bids, interpreted requirements, matched catalogue items, and generated qualified leads.
• Built enterprise information-retrieval and workflow agents using copilot studio, power automate, dataverse, and power apps.
• Implemented authentication, row- and field-level crm controls, and validation guardrails so outputs respected each employee's identity and permissions.

• Completed an intensive research-grade curriculum covering supervised learning, deep neural networks, probabilistic graphical models, generative ai, and reinforcement learning.
• Gained hands-on exposure to production-scale ml systems and model training and evaluation practices shared by senior amazon scientists and principal engineers.

• Conducted first-principles device simulation of lead-free perovskite solar cells using scaps-1d, varying absorber thickness, layer combinations, and back-contact work function to optimize efficiency.
• Co-authored and presented the paper 'enhancing efficiency of lead-free perovskite solar cell by varying thickness, layer combination and back contact work function' at ieee icecct 2024.

led 11 department heads and more than 100 executives across technical programming, workshops, and digital outreach for the student community.