Extra proof and context

Created a competitive coding platform at CubexO as an internal tool for developer hiring assessments and team skill-building. Developers join real-time multiplayer contests, solve algorithmic challenges in a Monaco-based browser editor with syntax highlighting, auto-completion, and multi-language support (Python, Go, JavaScript, C++). Code is executed in isolated Docker containers with resource limits (CPU, memory, time) via a Go-based judge service that returns verdicts in under 2 seconds. Features ELO-based ranking, live leaderboards via WebSocket, and post-contest analytics showing time-per-problem and solution patterns. Handles 200+ concurrent contestants with horizontal scaling on Kubernetes.

Built a cross-platform mobile health application for a CubexO healthcare client to replace their paper-based patient management workflow. Features include appointment scheduling with push notification reminders, medication tracking with dosage alerts, real-time vitals monitoring via Bluetooth wearable integration (heart rate, SpO2, blood pressure), and encrypted medical record storage with role-based access for patients, doctors, and admins. The NestJS backend handles HIPAA-compliant data encryption at rest and in transit, with PostgreSQL for structured records and Redis for session caching. Served 800+ active patients across 3 clinics within the first quarter of launch.

Designed and developed a decentralized voting platform as an internal R&D project at CubexO to explore Web3 governance use cases. Voters authenticate via MetaMask wallet signatures and cast encrypted ballots using zero-knowledge proofs generated with Circom/snarkjs — ensuring vote privacy while maintaining verifiability. Smart contracts deployed on Ethereum Sepolia handle vote tallying, quorum checks, and result finalization with on-chain event logs. Ballot metadata is stored on IPFS via Pinata, and a React frontend displays real-time results with animated progress bars. Handled 5,000+ test votes during internal hackathon with zero double-vote incidents.

Built a real-time object detection API at CubexO for a logistics client's warehouse automation. The system uses a custom-trained YOLOv8 model fine-tuned on 50K+ labeled images to detect and classify packages on conveyor belts with 96.3% mAP. Exposed inference through a FastAPI service with Redis-backed request queuing, achieving sub-50ms latency per frame at 30fps throughput. Added a drag-and-drop React UI for uploading images, reviewing detections with bounding box overlays, and retraining the model with corrected labels. Deployed as a serverless function on AWS Lambda with model weights cached in S3 for cold-start optimization. Monitoring with Prometheus tracks model drift and accuracy degradation over time.

Architected a production-grade infrastructure monitoring platform at CubexO to observe 15+ microservices running on AWS EKS. Aggregates metrics (Prometheus), logs (Loki), and traces (OpenTelemetry) into unified Grafana dashboards with custom alerting rules. Built a Go-based metrics collector using gRPC streaming that pushes data at sub-second intervals, and added anomaly detection for latency spikes using rolling-window z-score analysis. Integrated PagerDuty and Slack for incident routing with severity-based escalation policies. Reduced mean-time-to-detection from 12 minutes to under 90 seconds, serving 1,000+ active platform users.

Built a KPI analytics and demand forecasting dashboard during my time at Optico Solutions. The platform pulls sales, inventory, and customer data from multiple sources via REST APIs, runs feature engineering pipelines on 2M+ daily records, and visualises trends using interactive D3.js charts with drill-down capabilities. Implemented a strategy backtesting module where analysts can compare different demand models against historical data using Markowitz-style portfolio optimization. Automated weekly reporting workflows that previously took 8 hours of manual data preparation, and integrated Power BI exports for stakeholder presentations.

Developed a real-time collaborative whiteboard for a CubexO client's remote design team. Supports infinite canvas with shape tools, sticky notes, freehand drawing, and live cursors for up to 50 concurrent users per room. Built conflict-free editing using Yjs CRDTs synced over WebSocket rooms, with Redis pub/sub for cross-server message relay. Added version history with snapshot diffing so teams can revert to any previous board state. The frontend renders on HTML Canvas with a custom scene graph for 60fps pan/zoom even with 1000+ objects on screen. Used by 12 teams daily for sprint planning and design reviews.

Built an internal knowledge assistant for a client at CubexO that replaced their manual FAQ process. The system ingests company documents — PDFs, Notion exports, and internal wiki pages — into a Pinecone vector store, then uses a multi-step RAG pipeline (LangChain + GPT-4) to answer employee questions with cited sources. Handles 10K+ queries daily across 3 departments, reduced support ticket volume by 35%, and cut average response time from 4 hours to under 30 seconds. Deployed as a containerized FastAPI service behind an Nginx reverse proxy with JWT-based access control.