Amit Kumar.

Built a robust MLOps pipeline for a classification model, automating model retraining, testing, and deployment to a production environment using GitHub Actions and Docker. Achieved continuous delivery with zero-downtime updates. The codebase showcases production-grade MLOps best practices including automated model versioning with MLFlow, containerized deployment workflows, CI/CD pipeline configuration, and monitoring hooks for model performance drift detection.

Designed and built a full-stack web application to serve as a central learning and networking hub for the Data Science Festival community. Developed a content recommendation system to personalize learning paths for members. Implemented a mentorship-matching feature using a skills-based algorithm to connect junior data scientists with experienced mentors within the community. The repository demonstrates full-stack architecture with React/Next.js frontend, Node.js backend, and the collaborative filtering algorithm powering personalized content recommendations.

Developed an AI-native financial platform for novice investors, leveraging Large Language Models (LLMs) to simplify portfolio creation. Implemented an NLP front-end using prompt engineering techniques to translate natural language user goals (e.g., "invest in green tech with medium risk") into a structured, diversified investment profile. Integrated with real-time financial data APIs to track portfolio performance and execute trades. Explore the code to examine the prompt engineering framework, LLM response parsing logic, and the portfolio optimization algorithms that balance user intent with market data.

Defined the product vision and strategy for Zaro, an AI platform designed to solve the problem of fragmented marketing data for small businesses. Built scalable ETL pipelines to ingest and normalize semi-structured data from disparate sources, including Google Ads, Meta, and CSV uploads, using Python and Pandas. Implemented a recommendation engine to provide AI-based insights on campaign performance, budget allocation, and A/B testing strategies to optimize marketing ROI. Review the codebase to see the modular ETL architecture, API integration patterns, and the machine learning recommendation logic powering data-driven marketing decisions.

Identified an opportunity to make at-home health monitoring more accessible. Led the end-to-end product development of Palette, a mobile health app, from initial concept to MVP. Conducted user research to define the core value proposition and designed a simple UI to translate complex biomarker data into actionable insights for users. The live demo showcases the precision timer interface, real-time image processing pipeline, and classification model output rendered in an intuitive, clinically-accurate format—all built on Ionic/Angular with Python backend services.

An autonomous research pipeline built with LangGraph where four specialized AI agents — Supervisor, Researcher, Writer, and Critiquer — collaborate to research any topic and produce a structured, high-quality report. The system implements iterative quality revision loops where the Critiquer agent evaluates output and triggers rewrites until a quality threshold is met. Explore the codebase to see the LangGraph state machine, inter-agent messaging patterns, and how Together AI and Tavily are integrated for real-time web research and synthesis.