LLM Product Engineer / AI Chatbot Architect Needed for Diagnostic Sprint

We are building an emotional wellness app. It includes an AI chatbot guide, mood detection, and personalized recommendations of curated experiences such as practices, videos, and real-world actions. We are looking for an experienced LLM product engineer or AI app / chatbot architect to help diagnose and improve the current AI system. This is not a simple “write a better prompt” job. We need someone who can look at the full system, understand why the live app is not behaving correctly, and recommend the cleanest path forward. Current situation The app is already built and running in a test/live environment, but the AI experience is not yet good enough. The main issues: -Responses often feel generic, scripted, or shallow. -The AI asks too many questions instead of guiding the user or being proactive with recommendations. -It does not always follow through when the user answers or accepts an offer. -Recommendations do not always feel aligned with the user’s actual emotional state or need. -Some recommendations seem to default to calming content too often. -The live app still feels slow, with high perceived latency. -The system has accumulated prompt rules, routing logic, mood detection logic, recommendation tools, frontend rendering behavior, and database content that may not be cleanly organized. We need someone who can diagnose the system from both a technical and product-behavior perspective. What we are looking for You should have strong experience with: -LLM application architecture -Prompt systems and prompt organization -Tool calling / function calling -Conversation state and intent handling -Mood / emotion detection and sentiment analysis -Recommendation logic -Streaming responses and perceived latency -Evaluation systems for AI behavior -Debugging live LLM products, not just demos -Working with non-technical product/creative founders who have a strong vision Our stack includes React Native / Expo, Supabase, Vercel, OpenAI APIs / agents, and a content database. We are not looking for you to be the main frontend developer, but you should understand enough to diagnose whether latency or response issues are coming from the backend, frontend streaming, orchestration, prompt design, database retrieval, or app rendering. Initial scope: diagnostic sprint The first engagement will be a short paid diagnostic sprint. The goal is not to immediately rebuild everything. The goal is to understand the current system issues and recommend the fastest clean path to make the live app meaningfully better. During the sprint, we would like you to: 1. Review the current AI/orchestrator architecture. 2. Identify what controls response behavior, mood detection, recommendation selection, memory, tool usage, and frontend display. 3. Diagnose why the live app feels generic, question-heavy, slow, and why the output isn't matching our vision. 4. Identify whether the problems are caused by prompt design, routing, state management, recommendation logic, retrieval/database issues, frontend streaming, or deployment mismatch. 5. Recommend the smallest safe path to improve the live app. 6. Propose a cleaner architecture for future iteration. 7. Ideally ship one small, high-confidence improvement that is visible in the live app. The goal is not for us to tell you exactly what to change and you just execute it. We are not looking for someone to simply follow a list of implementation instructions. We are looking for someone who can understand the vision, evaluate the situation, explain the tradeoffs clearly, and propose the right solution. We need you to review the system, diagnose the root issues, and propose the cleanest path to make the live app meaningfully better. Product behavior we are aiming for The AI guide should not feel like a generic chatbot or therapist bot. It should: -Understand the emotional state beneath the user’s words. -Give meaningful observations, reframes, or direction. -Avoid unnecessary questions. -Follow through when the user accepts an offer. -Move the user toward an experience, practice, video, or action when appropriate. -Recommend content that feels specifically aligned to the user’s current state and perceived need. -Feel like a guide that helps create a meaningful emotional shift, not a generic chatbot that just keeps asking questions and the conversation going. Please include in your application 1. Examples of LLM products you have built or improved. 2. Your experience with tool calling, orchestration, and streaming. 3. How you would approach diagnosing a system like this. 4. How you distinguish a prompt problem from an architecture/state/retrieval problem. 5. One example of a time you improved the actual user experience of an AI product, not just benchmark performance. Please do not send a generic AI-generated proposal. We are specifically looking for someone who can think deeply about LLM product behavior and help us get this app to a usable early-user experience. ***DO NOT EMAIL OR CALL/TEXT ME. If you search for my info and try to reach out to me directly, you will be ignored and blocked.