Welprex - AI-Powered Vedic Astrology Chatbot
An experiment exploring the intersection of ancient wisdom and modern AI through agentic RAG pipelines.
Problem
Traditional Vedic astrology consultations are time-consuming and expensive, while most digital alternatives lack depth and personalization. I wanted to explore: Can we combine ancient Jyotish principles with modern AI to create an accessible, personalized astrology experience?
This started as a fun experiment to see how data from various systems—astronomical calculations, ancient texts, and user context—could be orchestrated through RAG (Retrieval-Augmented Generation) to present meaningful summaries that users could interact with and explore deeper.
Approach
Built an entertainment and wellness prediction platform positioned as an experiment rather than a predictive service. The key was creating a sophisticated data pipeline that intelligently decides what information to fetch and how to present it:
- Four-stage agentic pipeline: Classification → Data Fetching → Knowledge Integration → Response Generation
- Each stage makes decisions based on the user's query—whether to fetch birth chart data, which Vedic texts to retrieve, and how much context to include
- Implemented celebrity demo feature using publicly available birth charts to let users experience the service before sign-up, solving the "commitment before value" problem
- Privacy-first architecture with session-based data management (no permanent conversation storage)
Tech Stack
Backend
Django with Google OAuth authentication
Data Pipeline
MCP (Model Context Protocol) endpoints for real-time astrological calculations
RAG System
RAGFlow for selective retrieval from Vedic knowledge base (reduced data noise from 50KB+ to targeted retrieval)
AI Model
Claude for conversational responses integrated with retrieved context
Marketing
Google Ads targeting astrology-interested demographics, Instagram engagement, Midjourney for ad creatives
Analytics
Google Analytics with URL-pattern-based conversion funnels
Result
- Live platform with active users processing personalized astrological insights
- Conversion optimization: Implemented celebrity demo to address 2.3% visitor-to-chat conversion rate
- Cost-efficient: Selective data retrieval significantly improved response quality while reducing API costs
- Full-stack showcase: Demonstrates backend architecture, AI integration, and product thinking for freelance portfolio
What I Learned
Marketing & Positioning
Technical sophistication doesn't equal user adoption. Positioning matters—framing as "entertainment" navigated regulatory concerns while maintaining interest. Most critically: never ask for commitment before demonstrating value. The celebrity demo solved this fundamental UX problem.
Agentic Data Pipelines
Built a system that makes intelligent decisions about data flow—classifying user intent, selectively fetching astronomical data via MCP endpoints, retrieving relevant Vedic texts through RAG, and adapting the pipeline based on query type. This "decision-making data pipeline" was far more effective than a static retrieval approach.
Context Management
Learned to work within Claude's context window by dynamically switching context based on conversation flow, managing session state efficiently, and selectively including only relevant birth chart data and Vedic knowledge per query rather than loading everything upfront.
Product-Driven Development
Prioritized user experience over technical complexity. Simple, working solutions beat over-engineered features. Iterative feedback from mentors translated into structured GitHub tasks, balancing major architectural improvements with minor UI enhancements.