How emirkaanozdemr/yarinin-sehirleri-hackathon works
Project type and purpose
• This is a full-stack AI-powered marketplace prototype (“ServeSmart”) built for a hackathon, combining:
• A Streamlit-based web application (5 tabs for sign-up, selling, search, buying, analytics)
• Data-analysis pipelines (price-change mining, sales analysis)
• ML modules (time-series forecasting, text generation via external & local models)
• A static alternative HTML/CSS/JS interface as a backup or demo.
File structure
• Top level:
• Dockerfile, Makefile, requirements.txt → build & deployment
• Alternative Interface – Html,Css,Js/… → standalone front-end prototype
• Price Changes/… → CSV datasets & Jupyter notebooks for price-change mining
• Sales Analysis/… → Python scripts for sales metrics
• Text Generation/… → raw data + notebooks + small Flask apps for Llama & Gemma experiments + Gemini integration
• User Interface/… → Streamlit app (app.py), text-generation pipeline & prompt modules, model.h5 (LSTM), .env
• No central database directory—data lives as flat CSVs; user accounts likely managed in memory or via a simple store within Streamlit.
README for additional insights
• The Streamlit UI has 5 tabs: Create Account, Sell Product (with on-the-fly title/description improvement via Gemini API), Search Product, Buy Product, See Your Products + Analysis.
• Text generation uses an external Gemini‐1.5‐flash API; prior local fine‐tuned Llama/Gemma models are present but not in the final pipeline.
• A local LSTM model (model.h5) forecasts time-series data for the seller’s analytics tab.
Detail
• Every folder’s purpose as its own subcomponent if it contributes to runtime behavior (e.g., include “Sales Analysis/analysisV1.py” under Analytics Service).
• Document environment variables (.env) feeding secrets into the Text Generation and Purchase flows.