Using
- Gemini Deep Research
- Opencode
- Opus 4.6
- MiniMax 2.5
- Kimi K2.5
- Cloudflare MCP and Skills
- Agent Browser Skill
- GSD
- Desloppify
#1 Gemini Deep Research
Build a typescript engineering plan for this, it should include actual architecture decisions, libraries, tasks, frameworks and so on. Be opinionated. This will be handed off to a set of agents (claude) who will build this for us.
With scenario.pdf as an attachment
#2 Claude Opus 4.6 with gsd-roadmapper agent
Using docs/scenario.md and docs/engineering.md you should create a highly opinionated roadmap that can be handed off to various agents to build. All data to be provided should be syntetic which you can provide. High level roadmap could be design of the ETL,synth data, creation of the frontend and scenario/analysis tool, and QA testing you will have full access to cloudflare MCP in which you can deploy the application and skills like agent-browser where you can QA it. if you need anything else I can add that too
#3 The roadmap was created with 7 phases
- Phase 1: Synthetic Data Foundation & ETL
- Phase 2: Statistical Computation Engine
- Phase 3: Scenario Engine & Portfolio Adjustments
- Phase 4: Risk Analytics & REST API
- Phase 5: Frontend Dashboard
- Phase 6: Cloudflare Deployment
- Phase 7: QA & Hardening
Each with roughly 3-5 tasks. From here it is simply a task of running gsd:plan-phase 1, gsd:execute-phase 1 and then gsd:verify-work 1 for each of the 7 phases.
As these ran each phase took roughly 10-20 minutes in the background depending on how many agents they spawned to do the tasks (usually 3-5).
- AI generative frontends look terrible, I would specifically give it a few design systemd and UI kits like on https://designsystems.surf/ to be inspired with or use variant.ai
- Generative code sprawls really quickly, I would go through and use https://github.com/peteromallet/desloppify to clean it up
- This really struggled with deployment not entirely sure why, likely wasnt clearly enough documented on the cloudflare side