My Contribution: The "Gemini Foundry" Intelligence Layer & System Architect
I led the full-stack development and AI integration for Gemini Foundry, transforming a standard Next.js application into a sophisticated AI-powered startup accelerator. My primary contributions focused on strict structured outputs, system resilience, and financial analysis.
1. Core AI Engine & "Investor Persona"
Structured Intelligence: Engineered the integration with Google Gemini Pro 1.5, utilizing strict JSON Mode and Zod schemas to guarantee that the AI always returns parseable, structured data—essential for generating complex MVPs and financial models without hallucination.
Prompt Engineering: Designed the ruthless "Silicon Valley Investor" persona and "CFO Mode" system prompts, ensuring the AI delivers critical, high-value feedback rather than generic advice.
2. Specialized Features
CFO Mode (Financial Analysis): Built a dedicated financial modeling agent that generates 5-year projections, burn rates, and runway analysis. implemented Recharts to visualize this data into interactive, board-ready charts.
MVP Generator: Developed the logic to break down abstract startup ideas into concrete, phase-by-phase technical roadmaps and feature lists.
Security & Admin: Implemented a robust Admin Dashboard with RBAC (Role-Based Access Control) to monitor user sessions, ban abusive users, and audit live chat logs for quality assurance.
3. Technical Architecture & robustness
Next.js 15 + Turbopack: Migrated the core infrastructure to Next.js 15, optimizing for speed and SEO.
Resilient Build System: Solved critical deployment challenges on Vercel by implementing Lazy Initialization for Supabase clients and relaxing environment variable validation during the build phase (
lib/env.ts
&
lib/supabase.ts
), ensuring 100% build stability.
Authentication Proxy: Re-architected the authentication flow using a custom Proxy layer to securely synchronize server-side Supabase sessions with client-side Google OAuth.
For the "Challenges" Section:
"One of my biggest technical challenges was ensuring the Vercel build pipeline wouldn't crash due to missing environment secrets during static generation. I solved this by refactoring our Supabase clients to use lazy instantiation and fallback values, decoupling the build process from runtime configuration."
Log in or sign up for Devpost to join the conversation.