I built a production-grade RAG system that lets you chat with your documents using AI.
Here's everything I learned about building AI apps that actually work 🧵
1
16
I wrote a detailed breakdown of the architecture, challenges, and code.
If you're building AI apps, this might help:
[Dev.to dev.to/blessedefem/i-built-a…]
Open to questions! 👇
1
5
Tech Stack:
- Backend: FastAPI + PostgreSQL
- Vector DB: Pinecone
- LLM: Google Gemini 1.5 Pro
- Frontend: React + TypeScript
- Deployment: Railway + Vercel
@googledevs , @FastAPI , @pinecone
Today, I built a Code Review Agent making use of a runtime(Bun), library(Vercel AI SDK) and an LLM(Gemini). I also added commit message & markdown file generation tools. #ALX_AIDEV, @alx_africa, @jbarbier, @kalibetre
76
API-Aware AI: Generating client code.
I used the OpenAi spec as the API to create context for my AI in generating client code(casting a vote and getting poll results) for the App. #ALX_AIDEV, @alx_africa
20
I configured Gemini CLI to use Supabase MCP server to enable Schema-aware AI interaction with my database. I then prompted for schema -related question and query helpful for the app. #ALX_AIDEV, @alx_africa, @jbarbier, @kalibetre
40
I installed Context7 MCP on Gemini CLI,prompted it to explain & give example of the live poll results in my app,in line with the latest features in Next.js 14. Based on its output,I refactored my codebase using Trae's IDE AI. #ALX_AIDEV, @alx_africa, @jbarbier, @kalibetre
74
I used CodeRabbit to auto review a pull request on my Github and generated summary feedback. I pushed a follow-up commit.
#Automated code review
#ALX_AIDEV @alx_africa
1
34
Today, I made use of an AI-powered CLI tool, Gemini CLI, rather than IDE, Trae. I installed tool on Windows Terminal.
I prompted the tool to explain the code in a file in natural language, which it did just like the IDE does
#ALX_AIDEV, @alx_africa, @jbarbier, @kalibetre
66
Generating Documentation
Accurate Documentation saves developers lengthy time of deciphering intent of written codes, and even loss of context.
#ALX_AIDEV, @alx_africa, @jbarbier, @kalibetre
88
AI-Assisted security Review and Debugging.
Because errors are possible, security reviews are of utmost importance
#ALX_AIDEV @alx_africa, @jbarbier, @kalibetre
58
The main poll management API implementation, was subjected to optimization by the AI agent & afterwards reviewed.
Benefits include:
> Reduced code duplication
> Memory Efficiency, &
> Faster compilation & execution
#ALX_AIDEV, @alx_africa, @jbarbier , @kalibetre
94
Advanced IDE workflows(prompt linking and context anchor). Day 6 #polling app #ALX_AIDEV @alx_africa
15
