I love building with AI, and I especially love to scale harnesses and evals for agents. I have made numerous tools and environments before, all with varying use-cases and offerings. From RL environments (for LM certainty) all the way to powerful tools like an open-source Notion/Linear/Calendar alternative with a powerful agent that integrates with it and all your apps, I have significant experience when handling models, refining their harnesses to work as best as possible, and making unique frameworks best optimized for token efficiency, parallelism, and reliability.

If you want to see more, you can also go to my portfolio and check out my various blogs and project highlights.

This is one of my favorite developments (and is still a WIP), and I use it myself every single day to manage notes, tasks, and my calendar, all with the power of Noetic AI at the forefront. It offers:

• Open-source code: It can be ran locally, contributed to by others, and more.
• The best unification of tools ever: Noetic combines every single tracking/saving tool that millions use daily—Notion, Todoist, Linear, Google Calendar, ChatGPT, and more. It takes the place of all of them and then some, perfectly unifying and melding them into one seamless ecosystem.
• Powerful AI agency: The agent harness has been rigorously designed and tested to maximize search coverage, speed, and interoperability between features within Noetic as well as remote tools and MCP servers, together.
• And more: This is super powerful and offers nearly everything else offers, even going as far as including workspaces. You can invite users into your workspace, letting you contribute all simultaneously.

This is still a WIP, but this is one of my more complex projects. It offers a unique multi-agent orchastration system, that isolates important steps of the development workflow:

1. Planning: Taking the task at hand into account, reading the codebase rigorously, and drafting an in-depth plan once certain and ready.
2. Implementation: Implementing the plan created, enabling a dedicated/structured implementation approach.
3. Testing and debugging: Once done, it is time for more rigorous testing to account for all edge-cases and issues not previously caught. If/when any are found, they will be debugged to find the root causes; if they are found, they will be passed back onto the beginning of the loop, letting the cycle continue until all requirements are satisfied.

The goal of this is to ensure maximal hierarhy, letting each agent perform its task(s). This can work for hours, and when done, it can file a PR for you; followups to Athena Cloud Code can enforce it to iterate further as well if need be.

This is one of my more exciting little endeavors. I always wanted to use Claude Code with any model, and other "proxies" were either crap, or lacked the features I wanted most. This though, offers everything I ever could've asked for:

• Both Anthropic and OpenAI Completion formats.
• API key, provider, and model-level fallbacks, reducing error-rates drastically.
• Proper tool-calling, mutli-media support, and streaming functionality.

The coolest part as well, was that CrofAI, which has tens of thousands of users, deployed my Anthropic method in beta; thousands of users use this endpoint in production for the likes of Claude Code today, and it has shown to scale well thus far.

This was a small project I worked on, and it essentially offers a simple and modular FastAPI framework, enabling anyone to create custom agent frameworks; people can use this (just as I have for Athena Cloud Code), and create their own agents with server-side persistence. It offers auth pipelines, database logic, server-side and client-side chatting/tooling methods, and much more.

• Languages: Python, Typescript/JavaScript, Tailwind/CSS
• Frameworks: FastAPI, Django, Next.js, React, Node.js, Bun
• Platforms: Git/GitHub, Docker, Vercel, Railway, Google Cloud

• FastAPI and Next apps
• API design and interfacing
• Creating long-running agentic systems

• Creating agentic tools and interfaces for both local and production usage
• Developing, testing, and scaling RL/eval environments for language models
• Improving agentic flow, struction, and concision throughout long-running and complex tasks