@@ -38,16 +42,125 @@
Mehul Goel
- 
- 
- 
- 
+
+
+
+
+
+
-
-
-
-
+
+
+
+
+
+ ☑ Leading interactive labs for ~60 students, totalling 120 minutes, teaching fundamentals of coding, C, and datastructures.
+ ☑ Orchestrate office hours for personalized student support, focusing on largely theoretical discussions about C, alongside debugging sessions.
+ ☑ Working with professors to develop new material and problem sets to effectively challenge students in a collaborative and engaging format that enhances learning.
+
+
+ ☑ Developed Computer Vision implementation based off ResNet to segment internals of 36 different iPhone models
+ ☑ Implemented SORT algorithim for sementing parts on a convery belt moving at 5 m/s at real-time (> 60 FPS) with 96% IOU accuracy
+ ☑ Used YOLO V8 Image Classifcation to train a top-down model to detect screws from 3 differnt angles on 720p cameras on 36 models of iPhones, 12 models of iPads, and 4 models of Apple Watches. Led the creation of a demonstration video that was presented to Apple and leads of 100+ member lab.
+
+
+ ☑ Led a team of 10 in building out autonomous vehicle driving and passing, including position communication, implementing a Model Predictive Controller, and racing at 30 mph.
+ ☑ Innovating a LiDAR + Stereo Camera setup for perception of surrounding vehicles for autonomous zero aid passing.
+ ☑ Assisted in development of a new website to improve traffic, potential member interest, and promotion for sponsors.
+ ☑ Control security of 3 remote workstations, a variety of 3D printers that can be accessed by the 100+ members of the club.
+
+
+ ☑ Developed performance modeling software to assist in chip development with 97% accurate benchmarks on a variety of ML workloads (BERT, ResNet50, etc.).
+ ☑ Improved prior softwares hardware resource utilization by over 46% with a novel weighted round-robin load management solution.
+ ☑ Created a memory modeling software that modeled memory packets throughout the chip, and the variety of paths to identify bottlenecks in current designs.
+
+
+ ☑ Published and presented this paper to over 30 audience members, passing a rigourous approval period with 2 seperate peer-revision trials.
+ ☑ Built universal modeling software for variety of different silicons to test against 10+ ML Workloads, compatible with CPUs, GPUs, and even TPUs.
+ ☑ Modeled memory and performance accuracy of Google TPU and other hardware within 5% of lab-tested measurements.
+