Developed an AI-powered Marketing Campaign Automation platform that helps businesses create, manage, and optimize marketing campaigns across multiple channels with minimal manual effort. The solution leverages Generative AI and conversational AI to streamline campaign planning, content generation, audience targeting, and performance analysis. The platform enables marketing teams to launch campaigns faster while improving engagement and conversion rates. Inspired by modern AI-driven marketing approaches that automate campaign workflows and personalized customer interactions.
5
3
456
Vachanamrut AI Companion — AI Knowledge Assistant
Designed and developed an AI-powered conversational assistant that makes the teachings of the Vachanamrut accessible through natural-language interaction. The system allows users to ask questions in Gujarati or English and receive context-aware answers grounded in the Vachanamrut knowledge base.
Key Features
RAG-based knowledge retrieval
AI-powered conversational Q&A
Gujarati and English language support
Voice-based question input
Context-aware responses grounded in source content
Structured knowledge processing and retrieval
Google Gemini AI integration
Simple, accessible interface for spiritual knowledge discovery
Focus: Generative AI · RAG · LLM · Knowledge Assistant · Voice AI · Multilingual AI · Full-Stack Development
1
241
I designed and built Pristine's entire digital operations platform, from database architecture and role-based workflows to a production-deployed web suite and an AI analytics layer.
The challenge was replacing a ceramic processing plant's paper-and-spreadsheet production logs with a system that floor operators would actually use, while giving management the traceability and insight the paper trail never could. Every decision focused on data integrity, least-privilege security, and keeping the interface simple enough for shift operators to work quickly.
I built four services around a single PostgreSQL schema. The operator app handles the full production lifecycle: raw material intake, drying plant shifts, multi-section ball mill batch entry, quality control, and finished-goods inventory, with Excel export throughout and navigation that adapts to each of eight distinct plant roles. A modular Express and Prisma API enforces that access control at the route level. A separate React-Admin console keeps master data (mills, grades, recipes, suppliers, users) out of the operator's way.
The most important piece was traceability. I designed a quality control gate that sits between production output and inventory. QC records are created automatically when a batch completes, reviewed by an inspector, and only generate an inventory entry if quality passes. Every finished bag carries an auto-derived lot number encoding its production date, flow rate, monthly sequence, and bag count, so any outbound shipment can be traced back through inventory and QC to the exact mill batch that produced it.
On top of that I built an AI analytics agent that lets managers query the plant in plain English, architected around a zero-data-to-LLM guarantee. The model only ever sees the database schema and the question, never a row of production data. Five independent layers enforce it: two least-privilege Postgres roles separating read-only queries from insert-only auditing, a sqlglot AST guard that rejects any non-SELECT statement, a per-role table allowlist, the prompt design itself, and a full audit log of every interaction. Answers stream to the browser over Server-Sent Events through a LangGraph pipeline that routes, guards, plans the query, executes, then explains the result in business terms.
Tech Stack: React 19, TypeScript, Ant Design, TanStack Query, Tailwind CSS v4, Node.js, Express, Prisma, PostgreSQL, React-Admin, MUI, Python 3.12, FastAPI, LangGraph, OpenAI, AWS EC2, Nginx, Supabase, Vercel.
1
191
Project 4 — RAG Document Chatbot
Title: RAG Document Chatbot — LangChain · ChromaDB · Streamlit
Description: Fully local RAG application: PDF ingestion → 500-token semantic chunking → 73-chunk ChromaDB vector store → LangChain retrieval chain with source citations and multi-turn memory. Streamlit UI, no external API dependency.
Link: https://github.com/Krupa03/rag-document-chatbot
0
35