# mangowhois > Jihwan Ryu (류지환), who publishes as mangowhois: AI Agent Engineering · Backend & Infrastructure. I build agents whose decisions and actions can be traced and verified. I design and build agent execution boundaries, backends and cloud infrastructure, and study the conditions under which a model's judgment carries through to real execution and verification. This site is his engineering portfolio: 7 works (GEODE, Eco², Railshot, REODE, Kiki, Cotton, CRUMB), each on its own pages, one per chapter; a profile; PDF reports; and the documentation of GEODE, his agent harness. - Contact: jihwan.ryu@mangowhois.ai by mail; the profile links are listed below. - Editions: Korean is the source edition at https://mangowhois.ai/; English mirrors it at https://mangowhois.ai/en/; 30 more editions, translated from the English, sit at // (ja, zh-hans, zh-hant, ar, bg, cs, da, de, el, es, et, fi, fr, hr, hu, id, it, lt, lv, nl, no, pl, pt, ro, sk, sv, th, tr, uk, vi). Every link here is to the English edition. - Full text: https://mangowhois.ai/llms-full.txt has the English edition's text (the profile, every view of every work, the reports) in one Markdown file. ## Profile - [Jihwan Ryu](https://mangowhois.ai/en/#profile): AI Agent Engineering · Backend & Infrastructure; experience, education, awards, certifications and contact. - [GitHub](https://github.com/mangowhoiscloud) - [LinkedIn](https://www.linkedin.com/in/mangowhois/) - [YouTube](https://www.youtube.com/@mango_fr) - [Threads](https://www.threads.com/@mangowhois) - [Mail](mailto:jihwan.ryu@mangowhois.ai): jihwan.ryu@mangowhois.ai ## GEODE - [GEODE](https://mangowhois.ai/en/geode/): Autonomous Agent Harness · 2026.02.21 — present. Stack: Python, MCP, SQLite, Harbor. A local autonomous-agent harness you can hand research, file work and scheduled tasks. Role: Design, judgment, verification: Author · implementation: Claude · review: Codex. Opens on its Run loop chapter: The model picks the next action; the host keeps the execution boundary. - [GEODE: Overview](https://mangowhois.ai/en/geode/overview/): From observation to the next action. - [GEODE: Agentic engineering](https://mangowhois.ai/en/geode/development/): Implementations review sent back, research figures recomputed. - [GEODE: Completion](https://mangowhois.ai/en/geode/completion/): The model's stop proposal becomes completion only through verification. - [GEODE: Seed generation](https://mangowhois.ai/en/geode/seed/): Audit scenarios are produced through phases in a fixed order. - [GEODE: Policy search](https://mangowhois.ai/en/geode/sil/): Change one policy at a time; keep the measurement fixed. - [GEODE: Crucible](https://mangowhois.ai/en/geode/crucible/): Candidate acceptance and release authority are separate judgments. - [GEODE: Jev](https://mangowhois.ai/en/geode/jev/): Decision Offloading w. Jev. - [GEODE: Trajectories](https://mangowhois.ai/en/geode/trajectories/): From rollouts to evidence. ## Eco² - [Eco²](https://mangowhois.ai/en/eco2/): Backend · Cloud infrastructure · 2025.10.31 — 2026.02.10. Stack: Python, Kubernetes, Redis Streams, LangGraph. An AI assistant that understands how recycling works in Korea. Role: Team project · infrastructure and deployment foundation → domain backend → async processing, conversational agent, verification. Opens on its Cluster chapter: How the 24 nodes actually connect, one flow at a time. - [Eco²: Overview](https://mangowhois.ai/en/eco2/overview/): Recycling guidance from one photo and a conversation. - [Eco²: Agentic engineering](https://mangowhois.ai/en/eco2/development/): A fix kept apart from another agent, a review split by lens. - [Eco²: 24 nodes](https://mangowhois.ai/en/eco2/nodes/): The 24 nodes, counted by group. ## Railshot - [Railshot](https://mangowhois.ai/en/railshot/): AI deployment & operations platform · 2026.09.28 — 2026.10.04 · Completed. Stack: Python, MCP, IaC, AWS/GCP/OpenStack. Submit a project from the web screen, or from your own AI agent connected over MCP, and it carries through deployment file preparation, failure repair, image publication, environment rollout and public URL verification; after deployment the same screen shows history and logs and manages updates, stops, resumes and deletion. Role: Team project · my part: CI planning and implementation, CSP clusters and Ansible interface, operations, team code review and integration. Opens on its User flow chapter: The web screen and your own agent request deployment through the same product API. - [Railshot: Overview](https://mangowhois.ai/en/railshot/overview/): Deploy a locally built web app to prepared cloud and on-premises environments, then run it. - [Railshot: Agentic engineering](https://mangowhois.ai/en/railshot/development/): A handoff checked independently, two recoveries reconciled. - [Railshot: Service layout](https://mangowhois.ai/en/railshot/serving/): Operations and customer app clusters are separate, and the public path is kept apart from deployment control. ## REODE - [REODE](https://mangowhois.ai/en/reode/): Code migration · 2026.03 — 2026.04. Stack: Java/Spring, OpenRewrite, Maven, Docker. pinxlab AI R&D: a Java/Spring code-migration agent. Role: pinxlab AI R&D freelance · migration and verification system. Opens on its Repair loop chapter: A migration loop that keeps repair evidence apart from acceptance. - [REODE: Agentic engineering](https://mangowhois.ai/en/reode/development/): Making the repair agent read the configuration first. ## Kiki - [Kiki](https://mangowhois.ai/en/kiki/): Work signals and agent execution · 2026.04 — 2026.05. Stack: Paperclip, Slack, MCP. pinxlab AI R&D: a Paperclip plugin and a Slack collector that help a team run its AI agents. Role: pinxlab AI R&D freelance · workflow planning and development, investigating stalled execution and designing the recovery path. Opens on its Execution flow chapter: Between assignment and actual execution, a path to observe and recover. - [Kiki: Overview](https://mangowhois.ai/en/kiki/overview/): So that a proposal becomes the assignee's execution. - [Kiki: Agentic engineering](https://mangowhois.ai/en/kiki/development/): Not how many wake-ups, but whether a run started. ## Cotton - [Cotton](https://mangowhois.ai/en/cotton/): Translation workflow · 2026.05.06 — 2026.05.26. Stack: Hono, BullMQ, Drizzle ORM. pinxlab AI R&D: a game-dialogue translation workflow. Role: pinxlab AI R&D freelance · translation processing, judgment and review path. Opens on its Judgment flow chapter: Scores, retries and human review, each at its own boundary. - [Cotton: Agentic engineering](https://mangowhois.ai/en/cotton/development/): A race review found, stalled jobs returned by reason. ## CRUMB - [CRUMB](https://mangowhois.ai/en/crumb/): Generative Game Studio · Demo · 2026.05. Stack: TypeScript, Claude Code, Codex, Gemini CLI. CRUMB is a multi-agent studio harness for casual game prototyping. Role: Design and development · built with AI coding agents. Opens on its System chapter: The agents that build a game, kept apart from the evidence that judges it. - [CRUMB: Overview](https://mangowhois.ai/en/crumb/overview/): From a one-line game pitch to a playable draft, with the agents’ work in view and open to intervention. - [CRUMB: Agentic engineering](https://mangowhois.ai/en/crumb/development/): A silent exit made observable, a lost formatter caught at integration. ## GEODE product - [GEODE](https://mangowhois.ai/products/geode/?lang=en): Autonomous Agent Harness. Execute the task. Explore a better way to run it. GEODE chooses its next action from tool observations. An outer Experimental Loop searches scaffolds and evaluation scenarios, then manages candidate acceptance and reversion. - [GEODE docs](https://mangowhois.ai/products/geode/docs/?lang=en): the documentation, in English (Korean: https://mangowhois.ai/products/geode/docs/). - [GEODE docs for agents](https://mangowhois.ai/products/geode/llms.txt): every docs page with a one-line summary, in the llms.txt format. - [geode-agent on PyPI](https://pypi.org/project/geode-agent/): install with `uv tool install geode-agent`. ## Reports - [GEODE: Technical report (English PDF)](https://tech.mangowhois.ai/geode/report/en.pdf): 92 pages, 2026.09.15. - [GEODE: Technical report (Korean PDF)](https://tech.mangowhois.ai/geode/report/ko.pdf): 92 pages, 2026.09.23. - [GEODE: Decision Offloading w. Jev (English PDF)](https://tech.mangowhois.ai/jev/system-1-offloading/en.pdf): 81 pages, 2026.09.28. - [GEODE: Decision Offloading w. Jev (Korean PDF)](https://tech.mangowhois.ai/jev/system-1-offloading/ko.pdf): 81 pages, 2026.09.28. - [GEODE: JEVDB paper review (English PDF)](https://tech.mangowhois.ai/jev/db-review/en.pdf): 19 pages, 2026.10.11. - [GEODE: JEVDB paper review (English PPTX)](https://tech.mangowhois.ai/jev/db-review/en.pptx): 19 slides, 2026.10.11. - [Eco²: Technical report (English PDF)](https://tech.mangowhois.ai/eco2/report/en.pdf): 32 pages, 2026.08.27. - [Eco²: Technical report (Korean PDF)](https://tech.mangowhois.ai/eco2/report/ko.pdf): 32 pages, 2026.08.27. ## Source - [GEODE: Code](https://github.com/mangowhoiscloud/geode): public repository. - [GEODE: Eval artifacts](https://github.com/mangowhoiscloud/geode-eval-artifacts): public repository. - [Railshot: Code](https://github.com/Jasmin-Softbank/Railshot): public repository. - [CRUMB: Code](https://github.com/mangowhoiscloud/crumb): public repository. ## Optional - [Full text](https://mangowhois.ai/llms-full.txt): the English edition in one Markdown file. - [Korean edition](https://mangowhois.ai/): the source edition. - [Countries and regions](https://mangowhois.ai/choose-country-region/): every edition, by country. - [Sitemap](https://mangowhois.ai/sitemap.xml): every page of every edition.