I work at the intersection of AI systems, efficient vision models, agentic systems, and edge/distributed computing.
My current work spans lightweight Vision Transformer research, algorithm-evaluation infrastructure, safety-aware reliability agents, and reproducible optimization systems.
- Efficient Vision Models and Model Compression
- AI Systems and Distributed Execution
- Agentic AI
- Edge and Distributed Intelligence
- Reproducible Machine Learning Research
Status: Ongoing research
I am studying lightweight and hierarchical vision models through vHeat and Vision Transformer reproduction and analysis. Current work covers reproducible ImageNet validation, component- and stage-wise parameter/FLOP analysis, stage-aware FFN reduction, and surrogate-guided study of accuracy-efficiency trade-offs with frozen holdout evaluation.
Status: Established research artefact
A manuscript-oriented reproducibility package for capacity-feasible MEC task offloading, execution-node selection, and physical CPU allocation. It includes configuration-driven paired experiments across multiple baselines, controlled comparisons, ablations, sensitivity studies, statistical analysis, and raw evidence.
Status: Ongoing project
A safety-aware agentic workflow for diagnosing, planning, executing, verifying, and rolling back remediation actions in cloud-native systems. The project is being designed around reproducible fault scenarios and measurable diagnosis, recovery, MTTR, and safe-action outcomes.
Worked on a full-stack algorithm evaluation platform spanning FastAPI backend services, React/Vite frontend workflows, Kubernetes/Pod-based execution, and frontend-backend integration.
- Built and debugged run-management, structured-evaluator, correctness/quality-metric, artifact-discovery, and result-visualization workflows.
- Contributed to testing, CI, GitHub-integrated development, and iterative platform engineering across the evaluation lifecycle.
Status: Open-source software
A public Redis-backed distributed-execution foundation with a FastAPI control plane,
concurrent workers, atomic claiming, lifecycle controls, leases, retries, timeouts,
cancellation, heartbeats, observability, tests, and containerized deployment. Its
current main provides the systems foundation and does not claim completed formal
policy-performance conclusions.
Status: Open-source research software
A deterministic auditing tool for checking consistency across experiment configurations, raw runs, summaries, and reported research claims, with structured reports and CI-oriented validation.
- MEC Offloading Visualizer — a deterministic simulator for local, edge, and cloud baseline analysis.
- Supervised ML Foundations — reproducible supervised-learning and IoT predictive-maintenance learning and engineering practice.
- Efficient Vision Transformer and vHeat architecture research
- Model compression and surrogate-guided architecture search
- Agentic systems with measurable safety, diagnosis, and recovery behavior
- Reproducible AI-systems experimentation and research auditing
- Edge and distributed intelligence where it supports these research questions
ML / Research: PyTorch, scikit-learn, model evaluation, optimization, statistical analysis
AI Systems: FastAPI, Redis, Docker, Kubernetes, REST APIs
Frontend / Platform: React, Vite, visualization and evaluation interfaces
Engineering: Python, Git/GitHub, CI, testing, reproducible experiment pipelines