How kammili-kalyan/zepto-ai-ml-capstone works
The repository combines three distinct products: a book-data collection and SQL-analysis pipeline, Titanic exploratory/modeling notebooks, and an offline Zepto policy-support assistant. The diagrams preserve each module’s documented workflow and major stages. Some notebook internals and portions of the assistant graph were not sampled, so their internal wiring is shown only where supported by the README or available source.
Main entry points: graph.py Policy Retrieval · api.py Ask API · books.db Books Database · ingest.py Document Ingestion
Components 26
Book Data Pipeline
- Book Scraper
data_pipeline/scraper.py - Raw Books CSV
data_pipeline/books_raw.csv - Data Cleaning
data_pipeline/cleaned.py - Cleaned Books CSV
data_pipeline/books_cleaned.csv - SQLite Loader
data_pipeline/database.py - Books Database
data_pipeline/books.db - SQL Analysis
data_pipeline/queries.py - Pandas Comparison
data_pipeline/pandas_comparison.py
Titanic Analytics
- Titanic Dataset
analytics/titanic.csv - Exploratory Analysis
analytics/01_eda.ipynb - Model Training
analytics/02_modeling.ipynb - Trained Model
analytics/model_pipeline.joblib
Policy Assistant
- Ask API
support_assistant/api.py - Request Schema
support_assistant/schemas.py - RAG Workflow
support_assistant/graph.py - Intent Classification
support_assistant/graph.py - Policy Retrieval
support_assistant/graph.py - Direct Response
support_assistant/graph.py - Prompt Builder
support_assistant/prompt.py - Policy Documents
support_assistant/docs - Document Ingestion
support_assistant/ingest.py - Embedding Model
- Policy Vector Store
support_assistant/graph.py - Response Schema
support_assistant/schemas.py
Other components
- Books to Scrape
- Customer
Connections 27
- Books to Scrape → Book Scraper: provides pages
- Book Scraper → Raw Books CSV: writes records
- Raw Books CSV → Data Cleaning: loads data
- Data Cleaning → Cleaned Books CSV: writes cleaned data
- Cleaned Books CSV → SQLite Loader: loads data
- SQLite Loader → Books Database: writes tables
- Books Database → SQL Analysis: queries
- Books Database → Pandas Comparison: queries
- Titanic Dataset → Exploratory Analysis: analyzes
- Titanic Dataset → Model Training: trains on
- Model Training → Trained Model: saves model
- Customer → Ask API: asks question
- Ask API → Request Schema: validates input
- Ask API → RAG Workflow: calls ask_zepto
- RAG Workflow → Intent Classification: starts classification
- Intent Classification → Policy Retrieval: routes policy query
- Intent Classification → Direct Response: routes general query
- Policy Documents → Document Ingestion: provides documents
- Document Ingestion → Embedding Model: encodes documents
- Document Ingestion → Policy Vector Store: stores embeddings
- Policy Retrieval → Embedding Model: encodes query
- Policy Retrieval → Policy Vector Store: queries vectors
- Policy Retrieval → Prompt Builder: builds prompt
- Policy Retrieval → Response Schema: returns response fields
- Direct Response → Response Schema: returns response fields
- RAG Workflow → Ask API: returns response
- Ask API → Customer: sends answer