Senior Software & AI Engineer · AI Agent Platforms & Cloud Infrastructure
I work for a San Francisco company, building AI agent platforms, cloud infrastructure, and full-stack products.
I take systems from architecture to production: agent orchestration, tool calling, backend services, and Kubernetes deployments. My work connects Python and TypeScript applications with AWS / EKS, Terraform, and Helm.
9+ years building software, including Technical Team Lead at DEPT® and Technical Lead at Agnitu. I combine hands-on engineering with product decisions, client collaboration, and mentoring engineers.
🌍 Working remotely and building from wherever I land.
🛡️ Currently studying cybersecurity, with an interest in security for AI agent systems.
| Area | Focus |
|---|---|
| AI & agent platforms | Agent creation and deployment, orchestration, tool calling, MCP, RAG, LangChain, and LLM evaluation |
| Cloud & infrastructure | AWS, Amazon EKS, Kubernetes, Terraform, Helm, Docker, and CI/CD |
| Architecture & delivery | Application architecture, distributed systems, end-to-end product delivery, stakeholder demos, and technical mentoring |
AI & agent engineering
Cloud & platform engineering
Full-stack & product engineering
Architecture & technical leadership
More of my engineering background
- Services & integrations: REST APIs, FastAPI, Flask, tRPC, RabbitMQ, Amazon SQS, and C# / .NET Core.
- Data & cloud: PostgreSQL, SQL Server, MongoDB, Amazon S3 / RDS / ECS, and Google Cloud Platform.
- Product engineering: React Native, payments and Stripe integrations, CI/CD, Jenkins, Jest, and observability.
- Architecture & leadership: System design, technical delivery, product development, mentoring, and leading engineering teams.
A browser-based musical instrument driven by a connectome-based fly neural simulation. 166,700 neurons, 25.1 million connections, and a small interactive world that turns neural activity into sound.
TypeScript · React · Three.js · Web Audio
Explore the experiment ↗
Fine-tuning an 8B model on Apple Silicon for customer-support intent classification. The first adapter regressed; a lower learning rate and validation-based checkpoint selection brought accuracy to 88.89% across 297 evaluation cases.
Python · MLX · QLoRA · Evaluation
Read the findings and limitations ↗
From pixels to predictions: a small, readable PyTorch project for training image classifiers on MNIST, Fashion-MNIST, and CIFAR-10. Automatic dataset downloads, GPU selection, and saved checkpoints make it easy to run and change the experiment.
The first Fashion-MNIST check reached 66.31% test accuracy after one epoch on 2,048 training images using Apple MPS. This is an initial workflow check; the repository includes the settings and raw metrics so the result can be inspected.
Python · PyTorch · CNNs · Computer vision
Explore the code and training guide ↗
- Agents & AI infrastructure — orchestration, tools, and dependable execution.
- Cybersecurity — building my foundations and exploring security for agent systems.
- Learning by experiment — fine-tuning, reinforcement learning, and reproducible results.
Building useful AI. Exploring strange ideas. Sharing what I learn.





