/projects & intellectual-property
"Architecting robust infrastructure from concept to patent to production."
AI-Powered Real-Time System Monitoring and Optimization Platform
This patented system addresses the critical challenge of real-time server health monitoring in high-throughput computing environments. Traditional monitoring relies on static alert thresholds and post-mortem log review, leaving systems vulnerable to cascading micro-outages and silent memory leaks.
Our architecture integrates continuous OS-level metric telemetry with machine learning models: Isolation Forest for multi-dimensional anomaly detection and K-Nearest Neighbors (KNN) for workload pattern classification. Telemetry covers kernel thread counts, CPU usage spikes, memory allocations, disk I/O latency, and socket saturation.
The intelligent resource optimizer proactively flags aberrant compute patterns before downtime occurs, slashing manual log triage overhead by ~90% and reclaiming an estimated 15–20% compute waste.
Production Engineering Systems
Distributed Code Execution Sandbox
Engineered a fault-tolerant remote compilation and execution engine powered by Judge0 CE, Redis queueing, and AWS Auto Scaling Groups. Replaced single-server architecture to execute untrusted code in isolated, resource-constrained sandboxes.
Execution Latency Drop
Concurrent User Capacity
Self-Hosted Backend & Storage Platform
Architected a self-hosted backend platform eliminating cloud vendor lock-in. Routed Supabase PostgREST workloads into Amazon Aurora and migrated unstructured asset storage to Backblaze B2 via secure SSH tunneling with 100% API compatibility.
Cloud Storage Savings
Zero Vendor Lock-in
Research & Academic Systems
Multi-Objective Polypharmacy Optimization System
Datasets Integrated
- MIMIC-IIIClinical patient records
- DDInterDrug interactions
- SIDERSide effect resource
Polypharmacy — the concurrent administration of multiple medications — creates exponential combinatorial reaction spaces that human clinicians cannot exhaustively evaluate. This research formulates a reinforcement learning agent using Pareto-based multi-objective optimization to balance three conflicting clinical goals: maximizing therapeutic efficacy, minimizing adverse drug reaction (ADR) risk, and preserving patient tolerability.
By harmonizing three extensive clinical repositories (MIMIC-III, DDInter, and SIDER), our RL model explores non-dominated Pareto treatment frontiers, delivering reliable decision-support recommendations.