DEV Community

Cover image for EMOTIONAL Assistant
Obaidullah Siddiqui
Obaidullah Siddiqui

Posted on

EMOTIONAL Assistant

Launching v2 of my Emotion-Aware Chatbot: A lesson in backend deployment hell

I recently rebuilt and deployed version 2 of my AI chatbot, a system that uses LangGraph, FastAPI, and React to classify user intent, route queries to different agents, and perform Retrieval-Augmented Generation (RAG) with guardrails.

While building the features was intellectually stimulating, getting the backend to run live in the real world was an entirely different beast. I spent days fighting with deployment platforms. I tried Fly.io, but hit roadblocks with their persistent volume dashboards and 5-minute trial limits that kept suspending my machines. I tried Hugging Face Spaces, only to find out they had recently paywalled Docker deployments. I tried Koyeb, but their free tier no longer supports the persistent storage my SQLite and FAISS databases required.

I finally settled on Render. While it required adapting my Dockerfile to handle dynamic port binding, and forcing fresh Vercel builds to capture new environment variables, it finally worked.

For anyone deploying a full-stack AI app: the code is only 50 percent of the battle. Understanding platform limitations, memory constraints, and environment variables is the other 50 percent.

Frontend: emotinal-chat-frontend.vercel.app
Backend:https://emotional-logical-chatbot-2zrz.onrender.com/api/health

Top comments (0)