- Joined
- Jan 2026
- How they joined
- Directly
- Before Cognition
- AgentKitLead Software Engineer
- Status
- Current
- Quoted word for word from his 2026-10-28 LinkedIn post resharing the FrontierCode 2.0 launch: "Really grateful to be part of the FrontierCode 2.0 release alongside such an incredible team. 🙏
Thanks to Cognition and everyone behind the tasks, audits, and reviews!"Oct 28, 2026
- Kai is credited by name ("Kai (Tam Nhu) Tran") in the acknowledgements of Cognition's FrontierCode 2.0 benchmark post, in a list of contributors.Oct 27, 2026
- His own site describes the Cognition work as "Applied AI Engineer (Evals), Cognition Labs, 2026 - Present": authoring repository-grounded evaluation tasks and maintainer rubrics, designing programmatic and LLM-based grading criteria across architecture, correctness, instruction-following and code quality, and evaluating Devin, Claude and GPT against real open-source PRs for mergeability.2026
- He created and maintains CCS (Claude Code Switch, kaitranntt/ccs), a CLI for switching between Claude accounts, Gemini, Copilot and OpenRouter through an OAuth proxy. Repo created 2025-11-01; the GitHub API shows about 2,894 stars and 276 forks, with Kai as the top contributor (3,551 commits).Nov 1, 2025
- FrontierCode (v1 launched 2026-06-08) is Cognition's benchmark for whether a maintainer would actually merge an AI-written PR; tasks are built by 20+ open-source maintainers spending 40+ hours per task.Jun 8, 2026
- He is a core contributor to GoClaw (nextlevelbuilder/goclaw), "OpenClaw rebuilt in Go" with multi-tenant isolation, about 3,699 stars and 1,082 forks per the GitHub API (repo created 2026-02-22). He calls it "OpenClaw, but better" in a post.Feb 22, 2026
- His LinkedIn work history lists both "Applied AI Engineer (Evals) at Cognition (2026)" and "Applied AI Engineer at micro1 (2026)", and his site says he worked "as Applied AI Engineer with micro1 and Cognition". This is an inference, not confirmed.2026
- Lead Software Engineer at AgentKit (Nov 2025 - Aug 2026): his site says he architected the core platform and workflow engine of AgentKit, which it calls a commercial multi-agent AI development platform with 20+ workflow primitives. The site shows it with a ClaudeKit image, and his posts mention contributing to ClaudeKit.Nov 2025
- He holds the Cognition evals role at the same time as an AI Software Engineer job at Valsoft Corporation (since May 2026), building AI-enabled software for medical-affairs and information-request (IRMS) products in regulated pharma workflows.May 2026
- CCS launch post (2025-11-04): "AI-assisted coding shouldn't stop when your rate limits do." He built CCS so he could switch automatically between Claude Code, GLM 4.6 and Sonnet 4.5 when he hit API limits.Nov 4, 2025🚨 AI-assisted coding shouldn’t stop when your rate limits do. Hitting API limits mid-flow kills productivity - so I built CCS (Claude Code Switcher), a zero-config CLI tool that keeps your AI workflow smooth, fast, and cost-efficient. 💡 What it does: CCS automatically manages and switches between Claude Code, GLM 4.6, and Sonnet 4.5 - giving you seamless fallback between top-tier AI coding models without losing context or flow. 🔁 Why that matters: Each model excels differently: - GLM 4.6 → blazing-fast code synthesis and multilingual reasoning - Sonnet 4.5 → deep contextual understanding and long-context retention CCS lets you switch between them dynamically, optimizing for speed, accuracy, or context length — all from your terminal. 🧠 Key features: - Solve real world issues every top-tier coders facing. - Enterprise-level error handling - Cross-platform binaries for macOS, Linux, and Windows 🚀 Try it now: 🔗 https://lnkd.in/erNfeKha | 📘 https://ccs.kaitran.ca/ #AI #DeveloperTools #OpenSource #CLI #Productivity #DevOps #Automation #Claude #BuildInPublic
- At Orangead he was Software Engineer (Sep 2023 - Oct 2025): GenAI/RAG document Q&A, FastAPI services on AWS at 99.99% uptime with 38% lower compute cost, and YOLO + OpenCV vision pipelines cutting inference latency 80% on 28 edge devices. Before that he was Software Engineer (Jan-Sep 2023) and Blockchain Architect (May-Dec 2022) there, writing Solidity smart contracts.Sep 2023
- Public headline says "Applied AI Engineer @ Cognition | AI Software Engineer @ Valsoft | OSS Maintainer".
Authoring evaluation specs and grading rubrics for frontier AI coding agents • Author eval specs for autonomous coding-agent benchmarks on real software-engineering work. • Design programmatic + LLM-based grading criteria across architecture, correctness, instruction-following, and code-quality dimensions. • Iterate via PR-driven workflow: rubric hardening, calibration, and multi-run evaluation against frontier model families.
Developing AI-enabled software for regulated medical affairs and pharmaceutical workflows. • Develop AI-enabled software for medical affairs and information request management products. • Support pharmaceutical compliance workflows with reliable, documentation-aware AI features. • Contribute to specialized vertical-market software within Valsoft's long-term portfolio.
Building and contributing to open source AI infrastructure tools, used by developers worldwide. (not working FOR GitHub) • CCS (creator + maintainer): LLM CLI proxy unifying Claude, Gemini, Copilot, OpenRouter (300+ models) behind one OAuth layer. 2200+ stars, 190+ forks. TypeScript / Bun. • GoClaw (core contributor): Multi-tenant AI agent gateway in Go. 8-stage pipeline, pgvector memory, 11+ LLM providers, MCP bridge. 3000+ stars, 800+ forks.
Evaluating frontier coding agents and authoring grading rubrics for code-generation systems • Grade autonomous coding agents on real software-engineering tasks (architecture decisions, code quality, problem decomposition). • Author rubrics and evaluation specs for production agent training pipelines. • Cross-platform agent evaluation across multiple frontier model families.
Architected the core platform and workflow engine for AgentKit, establishing the technical foundation and guiding architecture through platform maturity. • Designed and architected Go and TypeScript platform services powering 20+ workflow primitives, deterministic agent orchestration, and developer APIs across hundreds of active sessions. • Guided the end-to-end architecture of production RAG systems (PostgreSQL/pgvector, Qdrant, Claude orchestration), deployed as Discord bots and embeddable web components. • Established evaluation standards and structured execution workflows that cut prompt iteration cycles by 60-70%, unlocking 4-5x productivity improvements for developers. • Drove technical strategy and platform roadmaps with product leadership, transitioning the codebase into a mature, maintainable foundation.
Architected scalable AI and edge-automation systems across cloud and on-device environments. • Integrated GenAI models for content generation and RAG-powered document Q&A workflows • Delivered 99.99% uptime across distributed FastAPI services on AWS while cutting compute cost 38%. • Built real-time vision pipelines with YOLO + OpenCV, reducing inference latency 80% on 28 edge devices. • Automated CI/CD with Ansible + Terraform, shrinking build time from 3 hours → 11 minutes.
Enhanced multi-platform deployment and automation for large-scale device operations. • Improved frontend response time 30% using Vue 3 + Vite. • Cut setup time 40% across Linux, Android, and iOS onboarding pipelines. • Boosted test coverage 3× and deployment stability via automated CI/CD workflows.
Designed blockchain-integrated AI systems improving scalability and data integrity. • Engineered Solidity smart contracts ensuring on-chain consistency across 10+ services. • Built Node.js + React modules reaching 150 TPS and reducing gas cost 22%. • Prototyped hybrid blockchain-AI architecture guiding future platform strategy.
Focus: Software Engineering, DevOps, AI/ML, and Systems Architecture Completed advanced coursework in Artificial Intelligence, Data Analytics and Web Services & Applications.
Drew myself the way my life actually looks these days: conducting a small army of AI agents, one hand on the keyboard, the other pointing at something that's probably on fire (yeah you, Antigravity, the RAM eater) Orchestrating agents turned out to be the easy part. Orchestrating my sleep schedule ... still open issue. Building CCS, 2k+ ★ and counting. Contributing to ClaudeKit and GoClaw (OpenClaw, but better). Thanks for believing in the chaos.
What happens when a solo developer solves their own AI workflow problems? Sometimes the solution finds its audience through organic growth driven by developer trust. This is that story... 🌱
🚨 AI-assisted coding shouldn’t stop when your rate limits do. Hitting API limits mid-flow kills productivity - so I built CCS (Claude Code Switcher), a zero-config CLI tool that keeps your AI workflow smooth, fast, and cost-efficient. 💡 What it does: CCS automatically manages and switches between Claude Code, GLM 4.6, and Sonnet 4.5 - giving you seamless fallback between top-tier AI coding models without losing context or flow. 🔁 Why that matters: Each model excels differently: - GLM 4.6 → blazing-fast code synthesis and multilingual reasoning - Sonnet 4.5 → deep contextual understanding and long-context retention CCS lets you switch between them dynamically, optimizing for speed, accuracy, or context length — all from your terminal. 🧠 Key features: - Solve real world issues every top-tier coders facing. - Enterprise-level error handling - Cross-platform binaries for macOS, Linux, and Windows 🚀 Try it now: 🔗 https://lnkd.in/erNfeKha | 📘 https://ccs.kaitran.ca/ #AI #DeveloperTools #OpenSource #CLI #Productivity #DevOps #Automation #Claude #BuildInPublic
