AI Engineering Manager at CDAI · Hands-on backend, AI & agentic systems architect · Creator of ModePot

I lead AI engineering teams and architect the systems they ship. I build high-performing teams and stay hands-on with scalable, robust backend, AI, and agentic systems designed to adapt as products, models, and infrastructure change.

ModePot is my open-source workshop for simplifying software, modernizing its foundations, optimizing performance, and inventing what comes next. I build frameworks, tools, engines, and games in my free time, with AI-native and agent-friendly approaches where they add real value.

• Dexpot: synchronous Python APIs with adaptive GIL and free-threaded execution.
• Intpot: define typed Python tools once and serve them through CLI, HTTP, or MCP, or convert existing applications between those interfaces.
• Summonpot: build API endpoints from simple typed contracts, combining exact application operations and bounded agent-owned decisions in one declaration.
• LifePot: an artificial-life game where AI proposes the ecology, deterministic code runs the world, and accepted changes are inspectable and replayable.
• RefPot: an experimental native C++ embedded database with Python CRUD and explicit transactions. SQL and ORM remain planned.
• UtilityPot: eleven focused tools for data, text, diagrams, and planning. Eight run locally in your browser, including screenshot OCR, image redaction, and spreadsheet comparison; three use bounded AI after you ask.
• SkillPot: practical agent workflows for building, researching, and presenting real work. In development, with no installable skills yet. The first planned workflow turns a working project into an evidence-backed demo.

I lead two separate cross-functional pods across a portfolio of more than six products. I operate well under pressure and tight timelines, bring structure to ambiguity, and make difficult tradeoffs without lowering engineering standards.

Results: In under six months, one pod reached 2.5× feature and product delivery velocity and the team launched three products. That delivery improvement led to responsibility for a second, separate pod. I also helped deliver a recommendation system that improved menu-item attachment rate 4×.

Previously, I built Haptiq's AI team and first AI infrastructure, recruited and led a seven-person team, and led technical delivery and client conversations for its first AI product, helping sell AI solutions and contributing to more than $2 million in new revenue.

Beyond ModePot, I have landed contributions in Hermes Agent and Superpowers:

• Hermes Agent: skill-install reliability, gstack Skills Hub integration, Kanban WebSocket disconnect handling, and session-list performance through lineage caching.
• Superpowers (dev branch): Windows/Copilot hook support and brainstorm-server shell-test reliability.

The linked pull-request histories also include work still under review.

My earlier research in physics and materials science produced 30 journal publications, one patent, and more than 1,000 citations. That work shaped the analytical, evidence-led mindset I bring to innovation, uncertainty, and difficult decisions.

• tugrul@modepot.io
• Personal site
• ModePot
• LinkedIn
• X
• Substack: AI, agentic systems, and the science behind them
• Passionately Curious: science for curious readers