Cvia — AI Resume Builder

Context

Most resume builders produce generic output poorly suited to tech roles. I wanted a tool that was opinionated about format, optimized for software engineering resumes specifically, and used AI not to write résumés from scratch but to sharpen existing bullet points into concise, impact-driven entries.

Constraints

• PDF generation from LaTeX is CPU-intensive and slow if done on the mobile device or a shared function — it needed to be isolated.
• The MVP had to ship in under 6 weeks to validate the idea before investing further.
• In-app purchases needed to work across both iOS and Android with consistent entitlement management.

Approach

I built the mobile app in Flutter with Appwrite for backend and auth. For PDF generation, I deployed a Dockerized FastAPI service wrapping a LaTeX compiler on Heroku — callable as a simple HTTP endpoint from the mobile app. Gemini AI was fine-tuned with prompt engineering to rewrite bullet points into the concise, metric-driven format common in strong software engineering résumés. RevenueCat handled cross-platform in-app purchase entitlements and pricing tiers.

Key decisions

• Isolated LaTeX-to-PDF in a separate Dockerized service on Heroku rather than running it inside a Supabase Edge Function or on-device — keeping generation fast, reliable, and cost-controlled.
• Used Gemini for bullet refinement rather than full generation, so the engineer still owned the content and the AI acted as an editor — matching the products opinionated philosophy.
• Chose RevenueCat over direct StoreKit/Google Billing integration to get unified entitlement logic across platforms without maintaining two purchase implementations.

Tradeoffs and risks

• Heroku introduces a cold-start delay on the free tier for the PDF service — accepted for MVP, with the plan to upgrade or migrate to a persistent service at scale.
• Appwrite was chosen for speed of setup over Supabase, trading some query flexibility for faster backend bootstrapping during the 5-week sprint.

Outcomes

• Shipped a working MVP to the App Store and Play Store in 5 weeks.
• Dockerized LaTeX pipeline reduced PDF generation operational costs by approximately 60% compared to an estimated on-demand cloud function approach.
• Gemini AI refinement produced approximately 30% improvement in resume quality scores based on early user feedback.
• In-app purchases with dynamic pricing tiers live on both iOS and Android via RevenueCat.

Artifacts

Technology