Devarajan S

Advanced microbiome causality analysis system for Stanford University combining Granger causality tests to identify causal relationships. Built AI validation pipeline with web scraping and custom deep research workflow to confirm microbiome cause-effect pairs.

Built an AI-powered Streamlit application that converts domain-specific research papers into structured datasets using NER pipelines. Leverages Spacy Toolkit and Groq API powered model evaluation to ensure high-quality entity extraction.

Developed an AI-powered Streamlit app using spaCy and scikit-learn to parse candidate CVs and match them against job descriptions. Automatically generates ATS-friendly suggestions and scoring.

Created a modular React + Flask productivity app with personalized study agents powered by LangGraph and CrewAI. Delivers dynamic, adaptive study plans and reminders.

Built a BART-based chatbot that ingests scientific papers on hydrogen storage and returns concise, on-demand summaries. Extracts and organizes key material properties and trends across multiple papers, boosting research accuracy and throughput.

Built an AI-powered automated calling system for Amrita Faridabad Hospital to collect patient feedback on care experience (PREMs) and health outcomes (PROMs) post-discharge. Enables interactive voice surveys with real-time analytics to improve healthcare service quality and patient satisfaction.

Developed an automated knowledge graph generation tool that fetches domain-specific data from Wikidata and Wikipedia using SPARQL queries, creating comprehensive relationship maps between topics and concepts. Built with async processing, MongoDB storage, and Redis caching to efficiently handle large-scale data retrieval and graph construction for educational content organization.

Built an advanced RAG system combining chain-of-thought reasoning (RARE) with intelligent multi-strategy retrieval capabilities for document-based question answering. Features dynamic knowledge base routing, multiple retrieval modes (chunks, metadata, content-based), and ChromaDB vector storage with Groq API integration for enhanced response generation.

Created a universal integration framework that bridges MCP tools, A2A agents, and LangChain components into unified AI workflows with seamless interoperability. Features automatic agent discovery, multi-threaded server management, and dynamic protocol conversion enabling cross-framework tool integration and workflow orchestration.

Developed a React Native application that captures skin lesion images and classifies them on-device using custom deep-learning models. Integrated real-time preprocessing and inference pipelines to deliver accurate melanoma vs. benign predictions.