I’m a machine learning researcher and senior Honors CS student at UC Berkeley.
I spent nearly a year deep in panoptic segmentation for satellite imagery, informal urban sprawl, and venture building around an idea advised by Grant Ricketts and Alejandro Maldonado, incubated at SkyDeck, recognized with Berkeley’s Changemaker Award at Big Ideas, and supported by AIEB. That chapter taught me a lot about execution under real-world constraints in Mexico.
I’m now rebuilding from first principles as a creator, developer, and leader.
Representative experiences are highlighted. This “Now” section uses verified timeline anchors only.
Alberto Hojel, Yutong Bai, Trevor Darrell, Amir Globerson, Amir Bar
We studied attention heads in visual prompting (MAEVQGAN) and used learned task representations to improve behavior, speed, and performance via targeted patching positions.
arXiv · OpenReview · GitHub
Alberto Hojel, Ryan Tabrizi, Heather Ding
Survey and implementation work on provably robust DNN methods using convex outer polytopes plus adversarial attack comparisons (FGSM vs box-constrained L-BFGS).
Alberto Hojel, Harrison Resnick, Owen Sleigh
Built and benchmarked a variational quantum classifier in Qiskit for synthetic binary classification with ZZ feature maps and Real Amplitudes ansatz circuits.
Alberto Hojel
Policy + technical analysis on Mexico’s property registry and how geospatial deep learning could help detect informal urban sprawl and support better infrastructure planning.
Energy and Resources Group
Worked on consumption-based GHG accounting and a modular climate action data portal for U.S. cities, combining diverse datasets into a practical decision tool.
Built cloud infra to stream Zoom transcripts into GPT-4 workflows for extracting insights and action items after meetings.
Built a geospatial AI approach for cadastral updates in Mexico using satellite segmentation and change detection. Admitted to SkyDeck Batch 15 and received Berkeley Changemaker recognition.
Elevator pitch · Deck · Award
Built an IoT-first industrial analytics platform for textile manufacturing: ESP32 nodes, MQTT pipelines, AWS IoT Core, DynamoDB, and real-time decision dashboards.
Implemented cloud data workflows and geospatial collection tooling for large-scale debt recovery operations, improving addressability, throughput, and collection performance.
Site · Slide deck · Impacts
Led discussion sections on ML, circuits, control, and signal processing while supporting a large course cohort with clear, structured technical teaching.
AI semantic search engine for course content · Yi Ma’s class breakdown