CV
EDUCATION
Master of Science in Electrical & Computer Engineering
Signal and Image Processing and Machine Learning
University of Michigan, Ann Arbor, MI · Expected May 2027
Bachelor of Science in Computer Engineering with Distinction
Purdue University, West Lafayette, IN · Dec 2024
EXPERIENCE
Robot Sensing AI System Intern
Industrial Technology Research Institute (ITRI) · May 2026 – Oct 2026 · Project page
- Trained a vision-based quadruped locomotion policy with PPO in NVIDIA Isaac Gym across 4,096 parallel environments (6 terrain types, 20 difficulty levels); in simulation, the policy learned to jump 0.9 m gaps and cross beam terrain with 0.3 m spacing
- Designed a CNN + GRU + MLP policy network fusing 289-dimensional visual and proprioceptive observations for terrain-aware control
- Distilled privileged teacher policies into a deployable vision-based student via dual-teacher distillation
- Closed the sim-to-real gap with domain randomization and sensor-noise injection, validating from Isaac Gym and Gazebo to the physical Unitree Go2
- Consistently climbed 15–18 cm stairs and 25–40 cm platform obstacles in Gazebo and on the real robot
- Wrote Bash scripts to automate routine tasks and used AI coding tools (Claude Code) in day-to-day development
Research Intern
University of Michigan Transportation Research Institute (UMTRI) · Mar 2026 – Jun 2026
- Built an AI-assisted HD-map lane editor combining agentic AI predictions with human review, correction, and QA for road-attribute mapping
- Developed a computer vision and geospatial pipeline that infers lane count, directionality, turn lanes, and median presence from aerial imagery and GIS data, automating HD-map attribute labeling
- Integrated SAM, DeepLabV3, and vision-language reasoning to validate roadway predictions against aerial imagery and GIS data, flagging errors before human review
RESEARCH EXPERIENCE
Research Assistant
PixAIL Lab, University of Michigan · Oct 2026 – Present
- Investigated possible causes of underestimated left atrial emptying, prioritized the likely factors, and designed validation tests using existing data
- Reviewed cardiac strain and twist metrics; identified through literature why circumference-based strain does not capture rotational deformation, supporting separate 3D torsion evaluation
- Proposed literature-backed fiber-aware regularization, size-independent twist metrics, and tagged-MRI benchmarks
Graduate Researcher
LASER (Life-cycle Assessment Synthesized with Ecosystems and Risk), University of Michigan · Aug 2025 – Sep 2026 · Project page
- Improved forest-biomass prediction R² from 0.66 (random-forest baseline) to 0.78 (RMSE ≈ 36.6 Mg/ha) by designing and training a multimodal FCN on Sentinel-2 imagery, AlphaEarth embeddings, and tabular data with PyTorch and Rasterio
- Fused tabular features into CNN encoders for multichannel image patches using residual MLPs and FiLM conditioning
- Built an end-to-end inference pipeline on the Great Lakes HPC GPU cluster (Slurm) to generate large-scale biomass maps for Michigan deciduous forest
- Performed GIS-based validation and error analysis to diagnose spatial artifacts and model failure modes
Undergraduate Researcher
ECELabs, Purdue University · Jun 2023 – Dec 2024
- Built and maintained an online FPGA experimentation platform used by 200+ students across 20+ FPGA boards, enabling remote hardware access and monitoring
- Designed a full-stack dashboard to track device performance, usage metrics, and system health
- Cut Diagram-to-SystemVerilog translation time 50× by replacing a Yosys-based conversion with a custom Python compiler that rebuilds netlists from schematic geometry, keeping user signal names and readable RTL
- Built a design-rule checker enforcing 13 rules on student schematics before synthesis (multidrive, combinational loops, bus-width mismatches, unconnected nets, PCF pin constraints), reporting all violations in one pass and cutting the fix-and-rerun cycle
- Developed VGA and audio hardware extensions for ICE40 FPGA using Pmod interfaces
Undergraduate Researcher
Autonomous Motorsport Purdue, Purdue University · Aug 2023 – Dec 2024 · Publications
- Cut manual labeling effort 90% with an automated YOLOv8 annotation pipeline that labeled 1,300 images
- Co-developed a CNN-based real-time track and lane detection system for autonomous racing under extreme conditions, 47× faster than YOLOv8 with 65% fewer parameters (ICLR 2024, Tiny Papers)
- Co-developed RoRaTrack, the first multi-camera road-racing dataset with pixel-level track annotations
- Co-designed RaceGAN (trained the generator and discriminator, proposed the post-processing step), a GAN-based model that outperformed 8 state-of-the-art baselines on racing scenarios (arXiv 2502.14068)
- Integrated perception models with the Purdue AI Racing vehicle and validated performance in the Indy Autonomous Challenge
TEACHING EXPERIENCE
Teaching Assistant
Department of Computer Science, Purdue University · Jan 2023 – Dec 2024
- Led weekly in-person C programming labs for 50+ students, held office hours, graded assignments, and organized exam review sessions
SELECTED PROJECTS
Senior Design Electrical Lead
Purdue University · Aug 2024 – Dec 2024 · Project page
- Led the electrical design for an autonomous vehicle, integrating custom PCBs for autonomous control and lane detection
- Designed multi-layer PCBs to manage power distribution and communication between the ESP32 microcontroller, H-bridge, and other components
- Developed ESP32 microcontroller firmware in C (including BLE for wireless laptop control) and a Python lane-detection algorithm, interfacing the Jetson Nano over UART with motor drivers for real-time control and detection
Processor Prototyping
Purdue University · Aug 2024 – Dec 2024 · Project page
- Designed and prototyped a 5-stage pipelined processor with forwarding, hazard detection, instruction/data caches, and a dual-core architecture in SystemVerilog on FPGA
- Implemented multicore and cache logic, including a bus-controller state machine enforcing snooping-based cache coherence
- Achieved up to 1.43× speedup on dual-threaded workloads with the dual-core design and characterized CPI under 0–10 cycle memory latency with a mergesort benchmark
Paris 2024 Central Planning Tool
The Data Mine, Purdue University · Aug 2022 – May 2023
- Partnered with Ray Ewry Sports Engineering Center to develop a central planning tool for the International Olympic Committee to coordinate logistics and scheduling
- Designed and implemented an ETL pipeline to process over 10,000 data entries using Python and SQL
- Developed a full-stack web application to allow partners and organizers to view and consolidate data for Paris 2024
PUBLICATIONS
- Ghosh, S., Chen, Y. H., Huang, C. H., Jameel, A. S. M. M., El Gamal, A., & Labi, S. (2024). Weighted branch aggregation based deep learning model for track detection in autonomous racing. In The Second Tiny Papers Track at ICLR 2024.
- Ghosh, S., Chen, Y. H., Huang, C. H., Jameel, A. S. M. M., Ho, C. C., El Gamal, A., & Labi, S. (2025). A Racing Dataset and Baseline Model for Track Detection in Autonomous Racing. arXiv preprint arXiv:2502.14068.
CONFERENCE PRESENTATIONS
- Ghosh, S., Chen, Y. H., Huang, C. H., Jameel, A. S. M. M., El Gamal, A., & Labi, S. (2024). Weighted branch aggregation based deep learning model for track detection in autonomous racing. Presented at the 12th International Conference on Learning Representations (ICLR 2024), May 7–11, 2024, Vienna, Austria.
- Ghosh, S., Chen, Y. H., Huang, C. H., Ho, C. C., & Labi, S. (2024). Purdue Vertically Integrated Program (VIP) for the Autonomous Motorsports Purdue Club. Presented at Next-Generation Transport Systems Conference (NGTS-4), Sep 2, 2024, West Lafayette, IN.
AWARDS AND HONORS
- Share with the World Award for Vertically Integrated Projects, Purdue University (2024)
- Electrical and Computer Engineering General Scholarship, Purdue University (2023)
- Dean’s List, Purdue University (2021–2024)
TECHNICAL SKILLS
- ML & Data: PyTorch, NumPy, Pandas, Rasterio, GIS (QGIS)
- Programming: Python, C, SQL, Bash, Julia, SystemVerilog, JavaScript, MATLAB
- Methods: Computer Vision, Deep Learning (CNNs, GANs), Segmentation, Reinforcement Learning (PPO)
- Systems & Hardware: Embedded ML, Jetson Nano, ESP32, FPGA, RTL Design, ASIC Development, Linux, Slurm/HPC
- Tools: Git, NVIDIA Isaac Gym, Gazebo, AI coding tools (Claude Code)
- Languages: English, Mandarin