Backend & AI systems engineer - shipped Yaari's infra end-to-end, built ComplianceIQ's RAG pipeline solo, idea to live.

medium

github

Devayani B • Peerlist

DescriptionComplianceIQ is an enterprise-grade RAG (Retrieval-Augmented Generation) assistant built to help legal and compliance teams query, analyze, and stay current on regulatory documents - without manually combing through hundreds of pages.The Problem Legal compliance teams spend hours manually cross-referencing policy documents, regulations, and internal guidelines. Existing tools either lack accuracy on domain-specific legal language or don't scale to enterprise document volumes.The Solution ComplianceIQ combines hybrid search, advanced retrieval techniques, and multi-provider LLM routing to deliver accurate, cited answers from large legal document sets - with full audit traceability for compliance-sensitive environments.Key Features<ul><li>Hybrid Search with RRF (Reciprocal Rank Fusion) - combines dense vector search and keyword search for higher retrieval accuracy</li><li>Cross-Encoder Reranking - re-scores retrieved passages for relevance before generation</li><li>HyDE (Hypothetical Document Embeddings) - improves retrieval on complex/ambiguous legal queries</li><li>Multi-Provider LLM Routing - automatic failover and key rotation across LLM providers for reliability</li><li>Google OAuth with Role-Based Access Control - secure, enterprise-ready authentication</li><li>SQLite Audit Trail - full query and document access logging for compliance requirements</li><li>Scheduled Re-Analysis - automated periodic document re-checks via APScheduler</li><li>Report Export - generate PDF/Word compliance reports directly from query sessions</li></ul>Tech Stack Built with Python, using Qdrant as the vector database, LangChain for document chunking and orchestration, and Streamlit for the interface. Deployed on Railway (backend) and Streamlit Community Cloud (frontend), with SQLite powering the audit and governance layer.StatusActively developed - built from local prototype through full cloud deployment

ComplianceIQ

Developed a smart email automation tool that customizes and sends personalized emails to HR professionals using their names, companies, and designations, fetched from an .xlsx sheet. The tool dynamically tailors content and automates the outreach process to increase placement opportunities and professional networking. Extracted data from Excel using Python Generated customized email content using AI logic Integrated email sending via SMTP with fail-safe retry mechanisms Enhanced productivity by automating a typically manual and repetitive process

AI-Powered HR Mailer

Developed a doctor appointment booking app using Node.js,Express.js,React.js, and MongoDB for efficient backend logic, frontend UI, and database management.Integrated Cloudinary for image management and utilized Tailwind CSS for responsive UI design, ensuring a user-friendly experience for booking appointments.

DocSpot

LegalLens is a State Level OpenAI buildathon qualified project which is an AI-driven platform designed to help rural citizens, low-literacy individuals, and first-time legal document users understand complicated legal contracts safely and confidently. The system extracts, analyzes, summarizes, and explains legal documents in Tamil, Hindi, and English, ensuring every person can make informed decisions before signing anything important.

LegalLens - AI-Powered Legal Document Simplif

A Cloud-based photo management app built with React.js, utilizing Cloudinary for secure image storage and retrieval. The app allows users to upload,organize, and view their photos seamlessly, offering an intuitive and responsive user interface.

PhotoCloudX

SnapGuard is an AI-integrated web application that detects and verifies image tampering using blockchain technology. The system uses OCR (Optical Character Recognition) to extract details from uploaded images, generates a SHA-256 hash, and stores it securely on a blockchain. Upon future uploads, SnapGuard compares the new image's hash with the original to verify authenticity, enabling secure document validation. 1.Integrated OCR to extract textual content from images (e.g., IDs, certificates) 2.Used SHA-256 cryptographic hashing to generate unique image fingerprints 3. Leveraged blockchain to store and verify tamper-proof records 4. Built with ReactJS for frontend and Node.js/Express.js for backend logic 5. Ensured file storage and verification flow was seamless and secure

SnapGuard – Image Tampering Detection & Verif

UConnect is a seamless video conferencing app built with React Native, Redux, and Agora SDK for high-quality real-time communication. It features crystal-clear audio/video with Krisp noise cancellation, real-time transcribing via Fireflies, and secure Firebase Auth. Backed by Node.js, MongoDB, AWS EC2, and Docker, it ensures scalable, reliable, and productive collaboration across devices.