Inspiration

Learning new things and implementing them is the only goal right now for me since I count this as my hobby. Me being a full-stack developer, the only thing that can fascinate me is other tech domains. Which turned out to be AI. This will be my first officially working project ever. So either way i win this hackathon or not, I actually Made It Work! and that makes me happy.

What it does

This Application works as a Feedback portal in which the customers can provide their valuable feedback and submit it. This Feedback then get stored in a TiDB-Serverless Cluster where its vector embeddings are stored side by side. The Admin can visit anytime and extract the valuable feedback or I would say constructive criticism towards the service or product provided by the Admin's firm. This is done is few easy steps: First- the admin pick a range of date between which he/she thinks the feedback would be necessary Second- Click on Generate Feedback and it's done. Now wait for few seconds till the Vector search does the job of retrieval and the LLM model tidy the feedback and return it to the admin as a PDF document.

How I built it

Frontend Framework : Nextjs Backend : Flask Serverless : TiDB Model : phi-3 mini 128k(quant-4_K_M)

Challenges I ran into

  1. Tring to query the retrieval data using a metadata which was of Date type (solution was to make the Date as a 8-Digit number of format YearMonthDate and perform basic integer operation on it for query)
  2. Deploying a Nextjs+Flask Web Application which is yet still an issue (trying to make use of Vercel, but sadly it not working out for anyone)

Burnout count = 7

Accomplishments that I'm proud of

Made my first fully functional AI powered Web Application :)

What I learned

How RAG works

Built With

  • flask
  • nextjs
  • ollama
  • shadcn
  • tidb
Share this project:

Updates