I build production AI — agents, pipelines, and the infrastructure that ships them.

An open-source, production-ready multi-agent orchestration framework for Python backends. Built to fill the gap between simple LLM scripts and real production deployments — where async, persistence, observability, and provider-agnostic routing are non-negotiable.

A two-sided marketplace platform for AI data annotation — think Upwork, but purpose-built for AI labs sourcing verified annotators and high-quality datasets. Built at Lingtec AI. I own the backend infrastructure: KYC, assessments, wallet, QA, and dataset delivery.

A practice-based leadership and power skills training platform powered by AI. I architected the core AI infrastructure — a voice-enabled tutor, an interactive AI avatar, and RAG pipelines grounded in curated learning content — with a sub-second end-to-end latency budget.

An extractive clinical summarization engine built on fine-tuned SciBERT. Built to solve information overload for doctors reading EMR notes — without the hallucination risk that makes generative LLMs dangerous in healthcare. Final ROUGE-1 score of 0.3782, a 43.75% improvement over baseline.

A deep learning hate speech detector built on RoBERTa — and a first-hand account of the biggest trap in machine learning: why a model with 95% accuracy can be completely, dangerously broken. Fixed with class-weighted loss, F1-based evaluation, and a preprocessing pipeline that treats emojis as semantic signal.

A secure, high-performance payment processing backend built on microservices. Engineered for resilience under traffic spikes, with strict idempotency, distributed tracing, and a real-time transaction monitoring dashboard used in production. Cut partner onboarding time by 40% and hit zero dropped transactions at peak load.

An AI-first customer support ticketing platform with automated triage, routing tickets intelligently across WhatsApp, email, and chat — without human intervention on first contact. Custom AI agents act as an automated L1 support tier, gated by a confidence-threshold system that escalates uncertain tickets instead of acting on them.