Sachin Gupta

PowerPoint Web - Performance Team

Leading development of an AI Browser Agent

Leading autonomous drone development for SUAS 2026. Founding Member: Spearheaded the club’s growth (specifically SWE/ML aspects) from inception to a nationally competitive drone team, leading initiatives in flight logic, autonomy, and simulations from zero to hero. Real-Time Object Detection and Localization • Setup a benchmarking (latency, throughput, recall, mean average precision) pipeline to objectively assess model performance • Fine tuned a YOLOv11 mode on a balloon + bucket dataset from Kaggle (mean average precision: 0.924 → 0.95 (+2.6%), recall: 0.749 → 0.793 (+4.4%)) • Used a TensorRT model format for a 2x latency speedup, achieving real-time low-latency performance when running on the NVIDIA Jetson Nano AI Chip • Employed data augmentation strategies to improve generalization • Tensorflow + OpenCV blob detection to identify and locate IR hotspots (representing potential wildfire locations) (first evaluated Laplacian of Gaussian (LoG), Difference of Gaussian (DoG), Determinant of Hessian (DoH) blob detection algorithms) • Object Triangulation (based on pixel points and altitude) (numpy for vector calculations) to determine the travel vector to reach the desired object that was detected • AI generated validation data to simulate and test performance under unique occurrences (of which adequate data is not available or hard to produce) • Leading the Software/ML sub-team of 20 members, and a founding member of the overall Drone Club (60 members) Simulation/Testing • Using Gazebo and Nvidia Isaac Sim to create a virtual (life-like) simulation of the drone and environment, to enable easy model testing Embedded • Connected the Flight Controller, Nvidia Jetson Nano AI Chip, Motors, Ground Station and more components together • C++ programming • SLAM, Lidar Dashboard • Using AWS, Docker, and React to create an interactive map dashboard representing the drones environment, boundaries, and flight path • Showcase the drones real-time video feed

Creating a world simulation to enable data driven insights and optimizations when putting Zipline in new cities

Optimizing ML inference pipelines for Shop App recommendations. Developing observability tools for clear understanding and analysis of each layer of the recommendation system, making it easier to troubleshoot "weird" recommendations.

AI-powered root cause analysis (RCA) and automated resolution • 🤖 Created an autonomous AI-agent test case solver that correlates Errors in NSP routing logs with Code Bugs and previous Jira Ticket issues/solutions for automated and instant network repairs • Developed a knowledge graph in Neo4j to model Nokia Standards' patent data and uncover relationships between various documents • Built an LLM-powered AI agent using a Retrieval-Augmented Generation (RAG) pipeline to generate well-informed natural language responses based on the knowledge graph • 📈 Utilized Deep Learning libraries like Tensorflow 2.0 and Keras to Predict Future Growth of 5G and 6G Networks via RNNs (Low MSE of 10.24) • Worked on a research paper about forecasted AI-based network traffic. The data came from conducting a comprehensive analysis of existing publications, papers, and articles relating to AI and 5G/6G networks. • Mentored by MIT PHD Fellow (Senior AI Research Specialist @ Nokia) and IoT Professor at Aalborg University (Denmark) (Head of Device Standards Research @ Nokia Bell Labs) in creating accurate Deep Learning models

Consumer level open-source software for improving access to industry-exceeding Cloud and ML services

Rocket scientist • Developed a low-latency Rocket WIFI protocol using C++ to enable on-rocket camera livestreaming down to ground equipment and wireless communication between on-rocket components • Used C++, Python, and OpenCV to combine the 4 camera angles into a singular 360 degree immersive video • STM32 and C++ embedded software development Stay Tuned! • CAN-RGX: Will test our payload by going on a low gravity flight experience. Allowing us to better understand how our payload and rocket overall behaves in a low gravity environment • Experiment tests the behavior of liquid metal electronics (3D liquid circuits injected into elastics/polymers) in space. These circuits would be flexible and self-healing, thus having a large application to rockets and especially astronaut suits

Creating a fully autonomous fighting robot • Designed C++ Algorithms to ensure fully autonomous reactions to surroundings and effective responses to opponents • Processed Arduino, Ultrasonic Sensor, and Infrared Sensor data by cleaning, scrubbing, and storing it for later use • At the International Competition: Won 4 Battles and Lost 2 • Check out a really cool clip of my SumoBot autonomously reacting to the opponent and performing a takedown maneuver! https://youtu.be/4TSukrEIZKE

Nokia NSP (Network Services Platform) - implementing features for Cloud Network automation, control, and optimization • Designed an OpenStack Resource dashboard for simple stack/VM/node information viewing and easy deletion of failed network software • Created a Network Infrastructure failure restoration protocol • Created Bash Scripts for efficient and automated network provisions: 98.3% reduction in manual input • Developing environments with T-API (Postman/Swagger) implementation to simplify existing network tools for easier operation