Overview
Code to Content is a SaaS tool built for solo developers, indie hackers, and startup founders who ship code regularly but rarely have time to write about it. It connects to a developer's GitHub account, pulls recent activity, and turns it into ready-to-post X content — without sounding like generic AI output.
Problem
Developers ship code constantly but rarely talk about it publicly. Writing good technical content takes time most developers don't have, and generic AI writing tools produce text that doesn't sound like the person who wrote the code. Indie hackers and startup founders lose visibility and audience growth because their work never gets communicated, even though the underlying work is interesting and worth sharing.
Solution
The system reads a developer's GitHub activity — commits, PRs, releases — and converts it into X posts using a structured prompt system. Output is built around per-user writing voice profiles, so generated content matches how the individual developer actually writes instead of defaulting to generic AI phrasing. Content generation is template-driven: different activity types (a feature merge, a release, a refactor) map to different content templates, and few-shot examples are injected into the prompt to anchor tone and structure.
Features
- GitHub activity ingestion (commits, PRs, repo changes)
- Writing voice profile system per user, built from sample posts
- Structured prompt templates mapped to activity type
- Few-shot example injection to reduce generic AI tone
- Manual copy-paste publishing workflow, with edit-before-post by design
My Role
Designed and built the full backend: GitHub OAuth integration and activity polling, the voice profile data model, the prompt template system, and the Gemini API integration that generates the final content. Owned the prompt engineering work end to end — this was the core product problem, not a side concern.
Architecture
The backend is a Node.js/TypeScript service backed by MongoDB. GitHub activity is pulled per user and normalized into a common event shape regardless of source (commit, PR, release). Each event is matched to a prompt template, then combined with the user's voice profile and a small set of few-shot examples before being sent to the Gemini API. Generated drafts are stored against the source event so a user can trace any post back to the commit or PR that produced it.
Technical Stack
Engineering Challenges
The hardest problem was preventing generic, interchangeable AI output. A single shared prompt produces flat, sample-looking text that reads the same for every user. This was solved by separating concerns in the prompt pipeline: voice profile data (tone, sentence length, vocabulary patterns) is layered on top of a structured template (what kind of content this is), and few-shot examples are selected per template type rather than reused globally. This turned one generic prompt into a per-user, context-aware generation pipeline that produces noticeably different output for different developers given the same underlying commit.
Impact
Live in production. Reduces the time developers spend turning code into shareable content from a manual writing task down to a generate-and-edit workflow — review a draft instead of starting from a blank page.