SpecForge

Inspiration

SpecForge was inspired by a common problem: developers often jump straight from a vague idea into code. That usually leads to messy features, unclear scope, and AI-generated output that is hard to actually build from. We wanted to create a tool that forces the planning step first, helping people turn rough ideas into clear product requirements before writing code.

What it does

SpecForge turns a vague app idea into a structured PRD through an AI-guided interview. The user describes their idea, the AI asks focused clarifying questions, and the app generates a formatted Product Requirements Document with goals, target users, core features, user stories, acceptance criteria, and out-of-scope items. Users can copy the PRD or download it as a markdown file.

How we built it

We built SpecForge with Next.js using the App Router. The frontend handles the guided interview flow, progress state, PRD preview, copy action, and markdown download. The backend uses Next.js API routes so API keys stay server-side. Gemini is used as the primary AI provider, with OpenAI configured as a fallback provider if Gemini fails.

The UI was styled with a quiet editorial design direction: off-white canvas, warm ink text, soft cards, pill buttons, and a clean document-focused layout.

Challenges we ran into

One challenge was designing the interview flow so it felt guided without becoming too long or rigid. We also had to make sure the app could generate a PRD even if the user chose to stop early with “Generate Now.”

Another challenge was handling AI provider reliability. To solve that, we added a fallback system: Gemini runs first, and OpenAI is tried if Gemini fails. We also moved all API calls into server-side routes so secret keys are not exposed in the browser.

Accomplishments that we're proud of

We’re proud that SpecForge turns a messy starting idea into a useful planning artifact in just a few minutes. The app is simple, focused, and actually usable: users can answer questions, generate a PRD, copy it, download it, and start over.

We’re also proud of the provider fallback system and the polished editorial UI, which makes the app feel more like a thoughtful planning workspace than a generic chatbot.

What we learned

We learned that the quality of an AI-generated document depends heavily on the quality of the questions asked first. A guided interview produces better output than a single prompt because it forces the user to clarify users, goals, features, constraints, and success metrics.

We also learned the importance of keeping API keys server-side, designing graceful failure paths, and building small product flows that are complete end to end.

What's next for SpecForge

Next, SpecForge could support multiple document types beyond PRDs, such as technical specs, build checklists, user flow diagrams, and launch plans. It could also add saved projects, editable PRD sections, export to Notion or GitHub, and team collaboration.

A future version could let users regenerate individual sections, compare different product directions, or turn the final PRD directly into implementation tasks.

Built With

• gemini
• next
• openai