DigitalOcean - Devpost Description

Inspiration

Most AI coding tools stop at code. They do not help users discover which idea is worth building, they do not show the decision process live, and they still leave validation and deployment work to the user.

We built DigitalOcean to close that gap: start from an idea, validate it with AI and research, then move all the way to a live deployed app on DigitalOcean.

What it does

DigitalOcean is an autonomous AI product workflow with three modes:

Zero-Prompt Start: one click launches idea discovery from YouTube, research-backed evaluation, competitive analysis, and a live Kanban of GO / NO-GO cards. The user chooses any GO card to build.
Evaluate: a specific idea goes through the Vibe Council, a six-agent review that scores feasibility, market signal, risk, and user value.
Brainstorm: the same system shifts into idea expansion and produces a structured concept brief instead of a build verdict.

The core product experience is the Zero-Prompt Kanban: discovery starts automatically, the board fills live, and a GO card can be built and deployed from the same workflow.

How we built it

DigitalOcean is a dual-runtime system:

web/ is a Next.js 16.1.7 frontend for Zero-Prompt, evaluation, brainstorm, and monitoring views.
agent/ is a Python 3.12.4 backend that combines Gradient ADK, FastAPI, and LangGraph.

The build pipeline follows a contract-first approach:

Idea refinement
Council review or Zero-Prompt evaluation
OpenAPI contract generation
Layered code generation
Validation (syntax, imports, build, contract checks)
Deploy plus health verification

This structure lets us keep the user-facing story simple while the underlying system remains observable and testable.

DigitalOcean usage

We currently document 13 DigitalOcean capabilities across Gradient AI and the deployment stack:

Gradient ADK
Knowledge Bases
Evaluations
Guardrails
Tracing
Multi-Agent Routing
A2A handoff
Serverless Inference
App Platform
Spaces
Image Generation
Agent Versioning
MCP Integration

The product also uses Managed PostgreSQL for session and lineage data.

Models in the current build

gemini-3.1-flash-lite-preview for Zero-Prompt discovery and lightweight brainstorming
claude-sonnet-4-6 for council analysis and cross-examination
gpt-5.4 for strategist synthesis and document generation
gpt-5.3-codex for code generation
gpt-5.2 for CI repair

Challenges we ran into

Keeping a dual-runtime system consistent across a streaming UI, FastAPI gateway, and ADK runtime
Moving from one-shot generation to contract-first, layered generation
Validating generated apps before deploy instead of trusting raw output
Handling SSE and long-running jobs without making the product feel opaque
Coordinating YouTube discovery, paper search, and competitive analysis inside one visible workflow

Accomplishments we're proud of

A public Zero-Prompt experience that shows discovery and ranking live on a Kanban board
Four public generated demo apps with working URLs at the time of the latest submission audit
A single product that covers discovery, evaluation, generation, validation, and deployment
A clearer split between the user-facing Zero-Prompt flow and the operational dashboard

What we learned

Reviewers trust consistent evidence more than large claim counts
Contract-first generation makes frontend and backend changes easier to keep in sync
A visible live workflow is easier to understand than a hidden batch job
Submission copy needs the same level of rigor as the codebase itself

What's next for DigitalOcean

Improve the handoff from GO card selection to deployed app preview
Expand reusable generation templates and evaluation criteria
Add stronger iteration loops after deployment feedback
Keep tightening the Zero-Prompt submission experience around a single clear story