I'm Sushruth Murakare, Software Engineer

• —Built and maintained insurance analytics dashboards tracking user segmentation data for clients including Farmer's Insurance, reducing load latency by ~20% through optimized state management.
• —Integrated generative AI (GPT-4) into a production platform to summarize complex tables and chart data; leveraged AI coding assistants to accelerate development, improve code quality, and explore creative solutions.
• —Collaborated with 5+ cross-functional stakeholders (product, design, QA, backend engineering) representing the front-end team, participating in code reviews and proposing improvements to architecture and practices.
• —Led a team of 10 interns as Scrum Master, tracking project progress in Jira, facilitating roadmap discussions, breaking down 20+ requirements into Jira tickets, and ensuring 100% sprint completion across 4 Agile iterations.

• —Owned end-to-end development of a React Native nutrition app, integrating AI-driven personalization and social media audio extraction using yt_dlp, processing 100+ audio clips into structured recipe entries.
• —Built and optimized 6+ backend APIs using Node.js and FastAPI, integrating third-party and internal REST APIs, and designing efficient relational database queries to support scalable application features.
• —Mentored 5+ teammates through code reviews, identified workflow inefficiencies across the team and proposed process improvements, and collaborated with the founder to define new features.

• —Designed, developed, tested, and deployed a full-stack application independently, building scalable REST APIs and optimizing database schemas for performant queries, while leveraging Docker and AWS for containerization and deployment.
• —Integrated AI using LangChain and Google Gemini to build resilient and maintainable Python backend services that summarized satellite imagery and related analytical data, delivering instant results through cached and pre-analyzed responses.
• —Built interactive MapLibre layers to visualize 200,000+ rows of satellite data across 50+ emitters, using custom icons and optimized rendering techniques to improve user experience and map clarity.
• —Reduced average page load time by ~40% by monitoring system health and proactively improving performance by fixing React re-renders, optimizing PostgreSQL schemas, and eliminating redundant API calls.

• —Migrated REST APIs to GraphQL, designing GraphQL schemas to reduce data transfer by up to 60% and support 4+ cross-functional teams, earning internal recognition.
• —Developed a real-time usage and balance tracking feature for telecom customers, building a Java backend service that automated live monitoring of data, call, and SMS usage, reducing support queries by ~15%.
• —Led development of a 5-screen scalable component library for a client in Kuwait, establishing consistent design patterns and a reusable front-end architecture that enabled users to add services and contributed to a team award.
• —Implemented a bottom navigator in a React Native mobile app, making navigation 3x faster.
• —Implemented Detox end-to-end testing and expanded unit test coverage across releases, accelerating development cycles, improving product reliability, and removing dependency on manual testing.

• —Collaborated directly with the founder to design and develop core platform features for an early-stage healthcare startup, helping shape product requirements and technical decisions.
• —Built and integrated backend APIs and developed the complete multi-step questionnaire UI, enabling seamless collaboration between dentists and 10+ insurance partners.
• —Implemented real-time validation, auto-save, and dynamic dashboards, reducing submission errors by ~30% and providing actionable insights from questionnaire data for healthcare providers.

The Colorado DMV website had no public API and was protected with CSRF tokens that made automation seem impossible. I reverse engineered the multi-step booking flow by watching network requests, figured out how the hidden tokens passed between steps, and built a Python scraper using aiohttp and asyncio that checks all 36 Colorado DMV offices at once for available appointments. Deployed on Render with a self-ping keep-alive so it stays awake on the free tier.

Built a Retrieval-Augmented Generation (RAG) system to answer questions about my professional profile using resume, LinkedIn data, and project documents. Used LangChain to extract, chunk, and embed data, storing vectors in ChromaDB and retrieving context via semantic similarity search. Developed and exposed the RAG pipeline through a FastAPI backend to serve low-latency, query-based responses. Deployed the service on AWS EC2, configuring public IP access and resolving CORS and mixed-content issues for seamless client-server communication. Click the `Ask me` button on the bottom right of the screen to check it out :)

Built autonomous browser automation tool using Playwright and Google Gemini AI that interprets natural language commands and executes multi-step web tasks by analyzing screenshots and HTML content at each stage. Implemented adaptive decision-making loop with configurable retry logic for error recovery and automated HTML report generation demonstrating production-ready agentic workflow design.

Researchers had 200,000+ rows of satellite emissions data across 50+ industrial emitter types but no way to explore it without writing code. I built the entire platform solo — FastAPI backend, React frontend with MapLibre for interactive visualization, LangChain and Google Gemini AI layer for natural language querying, PostgreSQL, Docker, and AWS deployment. Researchers can type a plain English question and get accurate answers backed by real satellite data in real time.

Designing and developing full stack web platform using React and Python to visualize and interact with geospatial datasets, structuring PostgreSQL database and building RESTful APIs for data processing. Extracted and organized 12 TB of geospatial data from legacy server, successfully processing 5 TB to date while implementing data migration workflows. Researched cloud platforms (AWS, GCP, Azure) and documented deployment architecture to support production hosting, storage, and scalability requirements.

Developed an access control system using Python’s face recognition library along with TensorFlow, Keras, NumPy, and Matplotlib to recognize faces even when partially covered, enhancing security for masked individuals. I evaluated machine learning models for mask detection, achieving 90% accuracy with VGG19 and 80% with MobileNetV2 and ResNet50. Additionally, I implemented the MaskTheFace script to detect facial features and apply masks to images, generating multiple variations for improved training data. This project was created during the pandemic to address the increased prevalence of mask-wearing.

Crack detection in construction materials is essential for ensuring structural safety and minimizing maintenance cost. Traditionally, methods such as manual inspection are time taking and have a room for human error thus often require specialized tools. This project proposes a machine learning based approach using Convolutional Neural Network (CNNs) to detect cracks in concrete surfaces through image analysis. Concrete was chosen as the initial materia due its widespread use. A dataset of 2GB containing labeled images of cracked and non cracked concrete was used which incorporated only real samples. The results suggest that deep learning models can effectively automate crack detection, offering a scalable alternative to manual methods.

Digi E Store is a deliberately bad-design UX/UI project that simulates a scam-like e-commerce platform to demonstrate how poor UX decisions and dark patterns frustrate users and break core design principles like Fitts’ Law, Jakob’s Law, Hick’s Law, and ethical design standards. By intentionally violating usability, accessibility, and visual hierarchy across multiple screens, the project highlights the real impact of bad interaction design. This project won the competition, recognizing its clear execution and strong understanding of UX laws by showcasing what not to do in design.

This study explores the design and impact of a kindergarten assistant robot, TODO, created to meet classroom needs. Drawing on participatory design principles and insights from interviews with stakeholders, TODO is envisioned as a dynamic, interactive, and emotionally intelligent robot. The research focuses on two primary factors that influence the robot’s effectiveness: its personality and conversational timing. By examining how these traits shape user perceptions, the study seeks to understand the role of politeness and engagement in fostering productive human-robot interactions. Using iterative design processes, grounded theory analysis, and controlled experiments, this research lays the groundwork for developing socially responsive robots that can assist teachers and enhance the learning experience for young students.

Designed and developed a full-stack web platform for Global Friends, a Colorado-based nonprofit supporting international students and workers transitioning to life in the USA. The platform features a public-facing site with event listings, a resource hub covering American culture and DMV guidance, and a student portal for scheduling airport rides, meal signups, and event RSVPs. Also built a volunteer and admin dashboard for managing requests, matching students with American host families through the Global Families Initiative, and tracking photo consent. Built with Next.js, React, Python, and FastAPI, with a Google Sheets MVP data layer designed for seamless migration to PostgreSQL at scale.

Related Coursework

Related Coursework