work
Things I've built
Two AI products running in production, designed and built end to end, plus selected machine-learning work.
Case studies
Products I designed, built and shipped — architecture through to production.
KubeTrace
kubetrace.net
An AI-driven Kubernetes testing, security, and reliability platform — the entire platform, built solo. It continuously validates clusters for security, reliability, and configuration drift — and pinpoints root cause in seconds.
- ▹MCP-based multi-tool integration with real-time, read-only cluster connect
- ▹AI root-cause investigation across multi-cluster, regulated production environments
- ▹Automated security, network, performance, storage & API test suites with compliance reports
Rubricate
rubricate.net
An agentic AI voice-interview and evaluation platform — the entire platform, built solo — that vets domain experts at scale with adaptive voice interviews backed by a full automated evaluation pipeline. The expert network behind Dataclap.
- ▹Agentic AI voice interviews that adapt to each candidate in real time
- ▹End-to-end automated evaluation pipeline with scoring rubrics
- ▹Automation workflows for sourcing, screening, and expert onboarding
Other projects
Machine-learning and application work, mostly open source.
Agentic RAG Chatbot
A local agent-driven RAG system that combines LLM capabilities with tool integrations to deliver accurate, contextual AI responses without relying on external APIs.
- ▹5-step reasoning pipeline for transparent, auditable reasoning
- ▹Local AI system with tool integrations for contextual responses
- ▹No dependency on external APIs
Accident Detection (Deep Learning)
A real-time camera monitoring pipeline to detect vehicle accidents and trigger alerts, built for real-time constraints, noise robustness, and edge deployment.
- ▹Real-time camera monitoring pipeline for accident detection
- ▹Object detection with temporal heuristics
- ▹Lightweight alerting backend for instant notifications
Fraud Detection Using GANs
A GAN-based augmentation pipeline and deep classifiers addressing extreme class imbalance in transaction data, with significant performance gains.
- ▹F1 score increased from 43% to 88% (↑45 pts)
- ▹Precision improved from 78% to 91%
- ▹Recall improved from 30% to 85%
Face Recognition Attendance
A Streamlit-based attendance system using facial recognition to automate student attendance tracking with an intuitive web interface.
- ▹Automated attendance tracking via facial recognition
- ▹Web interface for registration and marking
- ▹View and export attendance records easily
// next step
Want something like this built?
Tell me what you have in mind on a 15-minute call. I'll give you a straight read on scope, timeline and whether I'm the right person for it.