I lead product, engineering, and AI platform strategy at the point where ambiguity becomes execution.

Over 25 years in software, I have led and advised product, engineering, and strategy teams through platform bets, architecture risk, organizational complexity, and delivery pressure. My work is not only about building systems; it is about deciding which systems are worth building, aligning people around them, and carrying them into production.

I think like a product owner, operate like a strategist, and stay technical enough to know where the system will break. I care about teams and agents that can take responsibility in real codebases, not just look impressive in a terminal.

In 2026, while continuing my professional work, I made a deliberate turn back toward open source: to help public engineering teams, contribute to systems I respect, and make my judgment visible in the open. Occasionally loud in Turkish.

• 🪑 claude-roundtable — Multi-agent governance for Claude Code. Council deliberates, votes, dispatches, and enforces quality gates.
• ⏱️ leyla-scheduler — Durable, session-aware task scheduler in Rust. Jobs survive disconnects.
• 🔎 skill-hunter — Pre-execution layer that makes agents check for existing skills before building from scratch.
• 🛡️ promptguard — Audits prompts as behavioral contracts. Pre-write guard for agents that ship code.
• 🧠 recall-mcp — Shared brain for AI agents. SQLite-backed persistent memory over MCP.
• 👵 moooom-claude — 10 cultures of moms nagging your Claude Code to drink water and sit up straight.
• 🧰 pluginpool — Ten focused Claude Code plugins for everyday developer productivity. Each one is its own MIT-licensed repo: commit-narrator, pr-storyteller, test-gap, deps-doctor, env-lint, secret-guard, standup-gen, todo-harvest, flaky-detector, changelog-forge. 89 hermetic tests · Python 3 stdlib only.

External projects I'm actively shipping PRs into — open, merged, or in review. All public, all linkable.

Major OSS projects

• 🌐 google-gemini/gemini-cli — Google's official Gemini terminal agent. CLI ergonomics fix (session-ID rendering in resume messages).
• 🦙 ollama/ollama — local LLM runtime. Docs fixes (GPU FAQ link, Linux context-length config) and a ggml LoongArch quantization source PR.
• 🔎 elastic/elasticsearch — ESQL InvalidMappedField equality metadata fix.
• 📊 elastic/kibana — Fleet UI: use "Failed" label for failed integration status.
• 📦 elastic/elastic-package — system-test query for ignored fields.
• 🛰️ elastic/elastic-agent — os.Remove error guards in symlink/marker/enrollment cleanup paths.
• ☁️ cloudflare/wrangler-action — docs: add missing install and build steps to the canonical example.
• 🧱 strapi/strapi — four reliability fixes: review-workflows stages for unsaved locales, media-url refresh on replacement, null adminUserOwner in api-token service, and content-manager locale resolution for non-i18n delete paths.
• 🛒 medusajs/medusa — types fix: compound-word pluralization handling for uncountable suffixes.
• 🛡️ open-webui/open-webui — null-check in upload_image to prevent crashes when SSRF protection blocks a fetch.
• ☸️ kubestellar/console — store fix: use make([]T, 0) instead of var x []T for nil-safe JSON.
• 🛂 envoyproxy/ai-gateway — per-backend HTTPRoute cluster-name handling.
• 🤖 langgenius/dify — error class cleanup (removed unnecessary | None from typed error returns).
• 🧠 cline/cline — reset mistake count on user turn.
• 🤓 Aider-AI/aider — accept model-settings-file in benchmark mode.
• 🚦 maximhq/bifrost — return proper 400 for schema parse errors.
• 🔐 project-zot/zot — skip DynamoDB table creation when tables already exist.
• 🐙 oraios/serena — empty projects config-key handling.
• 🎨 shadcn-ui/ui — shrink-0 fix on base-luma SelectContent so the component sizes correctly inside flex layouts.
• ⚙️ actions/github-script — feature: working-directory input to change script execution directory.
• 🕷️ firecrawl/firecrawl — scraper fix: pass fullPage option to fire-engine request.
• 🪐 HeyPuter/puter — added a Havas Agentic OS MCP spine integration.
• ✨ refined-github/refined-github — fix on the most-downloaded GitHub UX-improvement extension.
• 🧩 gfazioli/mantine-split-pane — resizer fix: fire onResizing and onResizeEnd on double-click reset.

Claude Code / agent ecosystem

• 🧩 opencode ecosystem — opencode.nvim (edit-diff buffer cleanup, immediate interrupt commands), opencode-manager (repo-action tooltips, config-refresh model-form preservation), and anomalyco/opencode (run-root cwd, hyphenated MCP tool-id normalization).
• 🧰 openclaw ecosystem — openclaw/openclaw (pnpm baileys/libsignal git-deps, agent-runtime inheritance) and openclaw/wacli (NativeFlowMessage button extraction).
• 🛠️ MCP ecosystem — tosin2013/mcp-adr-analysis-server (issue templates), IN3PIRE/github-health-monitor-mcp (clearer rate-limit errors).
• 🪟 Claude Code tooling — YishenTu/claudian (recover from invalid settings JSON), mksglu/context-mode (atomic native-binary swap), slopus/happy (missing-agent-key recovery), steipete/CodexBar (Kimi K2 provider note).
• 🌀 Agent infra — manaflow-ai/cmux (install Gemini feed hook with BeforeTool), getpaseo/paseo, floatpane/matcha, recurram/recurram-zig (workflow-token permissions).

Heaviest external engagements

• 🔧 nexu-io/open-design — design-tooling daemon, largest external contribution surface (12+ merged + open PRs): zip import-failure surfacing, manual edit tabs, Claude smoke-test stream-json wiring, draw-overlay scroll, comment-marker alignment, picker hints, design-files panel grouping, memory-editor reveal, plugin-publish footer, and more.
• 🎬 nexu-io/looper — sequencer for long-running agent loops. PR-merge-blocker UI, pause/unpause commands, timing-sensitive test relaxations.

Skills distribution

Authored recsys-pipeline-architect and shepherded it into ~25 public skill / agent registries spanning the Anglophone, Chinese, and Turkish Claude communities: vercel-labs/skills, wshobson/agents, github/awesome-copilot, davila7/claude-code-templates, karanb192/awesome-claude-skills, VoltAgent/awesome-agent-skills, ComposioHQ/awesome-codex-skills, RoggeOhta/awesome-codex-cli, ikaijua/Awesome-AITools, Prat011/awesome-llm-skills, skillmatic-ai/awesome-agent-skills, heilcheng/awesome-agent-skills, kodustech/awesome-agent-skills, libukai/awesome-agent-skills (CN), laolaoshiren/claude-code-skills-zh (CN), travisvn/awesome-claude-skills, BehiSecc/awesome-claude-skills, ComposioHQ/awesome-claude-skills, rohitg00/awesome-claude-code-toolkit, jqueryscript/awesome-claude-code, GetBindu/awesome-claude-code-and-skills, spencerpauly/awesome-cursor-skills, shanraisshan/claude-code-best-practice, jeremylongshore/claude-code-plugins-plus-skills, Chat2AnyLLM/awesome-repo-configs, sickn33/antigravity-awesome-skills, affaan-m/everything-claude-code, gmh5225/awesome-skills.

• Leading or advising product, engineering, and strategy teams larger than 15 people for nearly 15 years.
• Leading product and engineering teams through ambiguity, delivery pressure, and architecture risk.
• Advising multi-disciplinary teams where product direction, technical strategy, and execution discipline have to move together.
• Turning fuzzy product direction into executable systems.
• Seeing the strategy, user workflow, and failure mode in the same frame.
• Choosing the next useful move when everyone else is still debating abstractions.
• Raising the bar without slowing the room down.
• Turning vague bug reports into concrete failing cases.
• Finding the smallest patch that a maintainer can actually review.
• Building guardrails so agents do not silently improvise their way into production.
• Staying with the boring part: CI, review comments, rework, and merge readiness.

I come from a mathematical way of thinking: definitions first, invariants second, implementation third. My research line connects product strategy with formal systems for AI agents:

• TeserracT — policy-constrained stochastic control, type-level governance, and market economics for multi-tenant agent platforms.
• Miray Tesseractic Mathematics — a semantic mathematics program for LLM meaning across geometry, energy, topology, type theory, memory, and human impact.
• Gated Persona Intelligence — stability-controlled persona memory for LLM systems; entropy gates, sigmoid smoothing, hysteresis, and deterministic cognitive state engines that reduce drift in long-running agent interactions.

That background shapes how I design products: formal enough to reason about, practical enough to ship.

The interesting failures in agent systems are not capability gaps — they are accountability gaps. Agents do not push back on scope. Memories drift into summaries. Self-correction is theatre without external validators. I write infrastructure that turns those soft failures into hard ones: a council that has to vote, a scheduler that has to resume, a guard that has to read the contract before the agent ships.

The goal is simple and hard: agents that can survive contact with real repositories, real tests, and real maintainers.

Product leadership · Technical strategy · Agent systems · Rust · Go · TypeScript / Next.js · Python · .NET / C# · SQLite · MCP · LLM evaluation

• 𝕏 @TuracTheThinker
• ✍️ Dev.to · TuracTheThinker
• 💼 LinkedIn

"Code is universal. Some commit messages are in Turkish. Make peace with it."