MLOps Engineer and Backend Developer bridging the gap between AI research and production infrastructure. I build systems that are deployable, scalable, observable, and—most importantly—reliable.

I take AI models from Jupyter notebooks and turn them into production services that handle real traffic. With a strong foundation in backend engineering and system design, I focus on the infrastructure that makes AI useful in the real world.

My work centers on:

• Model Serving & Optimization: Deploying models with low latency and high throughput (FastAPI, Triton, BentoML)
• Orchestration: Managing containerized workloads on Kubernetes with autoscaling and GPU scheduling
• MLOps Pipelines: CI/CD for models, experiment tracking, and versioned deployments (MLflow, Kubeflow)
• Observability: Structured logging, metrics, and alerting to ensure system health (Prometheus, Grafana)
• AI Agent Orchestration: Building multi-agent systems with human-in-the-loop safety checkpoints (LangGraph)

Philosophy: AI systems should be boring but bulletproof. They should work predictably, fail gracefully, and integrate seamlessly into existing workflows.

(Qwen Global AI Hackathon 2026 - Track 4)

Problem: On-call engineers waste 10+ minutes just gathering context during incidents.

Solution: An AI-powered SRE assistant that handles the first 5 minutes of an incident:

• Fetches Kubernetes logs and metrics in real-time
• Uses Qwen API to hypothesize root cause in < 3 seconds
• Generates safe mitigation commands with an is_destructive safety flag
• Drafts post-mortem reports automatically

Tech Stack: FastAPI, Qwen API, Kubernetes Python Client, Redis, Docker, Alibaba Cloud ECS

Live Demo | GitHub | Devpost

(Production RAG for Internal Documentation)

Problem: Teams spend hours searching wikis, PDFs, and Slack history for answers.

Solution: A retrieval-augmented generation system with:

• Document ingestion and semantic chunking
• Vector database storage (Pinecone/Chroma)
• Natural language Q&A with source citations
• Human feedback loop for continuous quality improvement

Tech Stack: FastAPI, LangChain, Qwen API, Vector DB, Redis, Docker, AWS S3

GitHub

(Autonomous Research Pipeline)

Problem: Complex tasks require multiple specialized steps (research, extraction, summarization).

Solution: A LangGraph-based system with:

• 3 specialized agents (Researcher, Extractor, Summarizer)
• Human-in-the-loop checkpoints for validation
• Persistent state management across sessions
• Async task queue for long-running jobs (Celery + Redis)

Tech Stack: LangGraph, FastAPI, Celery, Redis, PostgreSQL, Docker Compose

GitHub

Every system is designed for reliability—not just functionality. I evaluate latency, cost, failure modes, and rollback strategies before writing a single line of code.

Infrastructure is defined in code (Terraform, Helm, Docker Compose). Manual clicks in cloud consoles are a liability.

If it can't be monitored, it's not production-ready. Every system exposes metrics, structured logs, and traces.

AI systems need guardrails. I enforce human approval, destructive-command flags, rate limiting, and graceful fallbacks to prevent unintended harm.

AI should be intelligent and economical. I optimize token usage, use cheaper models for simple tasks, and cache aggressively.

I'm actively seeking MLOps Engineer, AI Platform Engineer, and Production AI Systems roles where I can build the infrastructure that makes AI reliable, scalable, and impactful.

If you're building AI infrastructure or need someone to take models from notebooks to production, let's talk.

⭐ Built with ❤️ for production-grade AI.