I am looking for a skilled Full Stack Developer/AI Engineer for a long-term partnership. We are building a high-end AI coaching application (a sophisticated Gemini Wrapper) and have a roadmap for several future applications.
I have a background in "vibe coding" (using AI to accelerate development) and have already built the initial infrastructure, but I need an expert to help troubleshoot a specific API block, finalize the app, and handle future scaling.
The Current Project Status:
We are building a secure RAG web app using Google Antigravity / Firebase. I have a complete Technical Production Plan (generated by Gemini 3.0 Pro) and an Execution Plan available.
You will be able to take over the existing codebase. Here is exactly where the project stands:
❖ Phase 1 (Infrastructure & UI): Complete. React (Vite) + Tailwind (Liquid Glass design) is live. Firebase Auth and Hosting are set up.
❖ Phase 2 (RAG Pipeline): Complete. Cloud Functions successfully parse .vtt uploads. Firestore Vector Search is indexed. Vertex AI (Gecko) is successfully generating embeddings.
❖ Phase 3 (Chat Engine): Currently Blocked. The logic for Context Retrieval is written, but we are hitting a persistent 404 Not Found (Publisher Model) error when the Cloud Function tries to call the Gemini API via Vertex AI.
❖ Phase 4 (Integrations): SMTP and Zapier integrations are pending.
The Project outlined:
We are building a standalone "AI Coach" for our community. This project moves our current internal CustomGPT solution into a secure, standalone software platform with its own login system. It is a RAG-powered consulting engine that allows members to "talk" to our entire library of training videos and documents.
The goal is to give every member a 24/7 consultant that answers questions using only our verified strategies, citing specific timestamps in our video training.
The Architecture (How it Works)
We have designed the app using a biological metaphor to keep the structure clear:
1. The "Brain" (Intelligence & RAG)
❖ Core Intelligence: We use Gemini 3 Pro (via Vertex AI) for high-level reasoning.
❖ Memory: All our training videos (.vtt transcripts) and documents (.pdf/.docx) are stored in Firestore.
❖ The Workflow: When a user asks a question, the system searches our vector database for relevant advice, retrieves the exact video timestamp, and answers the user while citing the source material.
2. The "Body" (Frontend & Design)
❖ The Vibe: The UI follows a "Liquid Glass" aesthetic. Think Apple-like minimalism—translucent white backgrounds, soft blurs (backdrop-filter), and subtle light blue accents. It feels premium and calm.
❖ The Interface: It is a secure web app (React + Vite). The sidebar (old chats) is collapsible for mobile, and the chat interface is "sticky" at the bottom, just like a modern messaging app.
3. The "Nervous System" (Automation & Logic)
❖ Access Control: Access is strictly for members. We use Zapier to sync with GoHighLevel. When a member joins or cancels in GHL, Zapier triggers a Cloud Function to grant or revoke their access immediately.
❖ Safety Limits: To control costs and prevent abuse, we have a hard limit of 50 messages per user/day.
What I Need From You:
❖ Troubleshoot the "Brain": Immediate priority is fixing the Vertex AI/Cloud Function permission or region issue causing the 404 error so the frontend can chat with the model.
❖ Finish the Build: Complete the SMTP integration, finalize the chat interface (citation logic), and polish the UI.
❖ Long-Term Collaboration: Once this app is stable, we have a queue of other AI-driven tools to build.
Technical Stack & Requirements:
❖ Core: React (Vite), Tailwind CSS, Firebase (Functions, Firestore, Auth).
❖ AI/LLM: Vertex AI, Gemini 1.5/3.0 Pro, Firestore Vector Search.
❖ Experience: You must have experience building "Wrappers" or RAG applications.
❖ Workflow: We have a set plan, but I am open to your suggestions if you see a better technical approach.
Why work with me?
❖ Clear Communication: You will communicate directly with me. I understand the tech stack and the "vibe coding" workflow, so we won't waste time on non-technical explanations.
❖ Preparedness: The project files, vector database setup, and system prompts are ready to go.
❖ Future Work: This is not a one-off gig. I am looking for a lead developer for my portfolio of apps.
Please apply with a brief summary of your experience with Firebase and Vertex AI, and let me know if you are comfortable debugging the specific 404 error mentioned above or if you want to start from scratch.
Screening Questions (Optional but Recommended):
Have you deployed a RAG application using Firestore Vector Search before?
Based on the description, what is your initial hypothesis for why a Cloud Function would return a 404 Publisher Model error despite the project ID and region appearing correct?
Are you comfortable working with codebases that have been partially generated via AI ("Vibe Coding")?