Projects
A language-driven pick-and-place system built around a $400 SO-101 arm (~$600–1,000 with depth camera and edge compute): Whisper, YOLO-World, SmolVLA, and a GQ-CNN verify-before-act gate, reaching 96% task success.
A vision-guided robot that uses YOLO detection to find and collect softballs on a full field, advised from concept to working hardware — covered by the Harvard Gazette.
A zero-install runtime digital twin for CNC process monitoring: it streams machine telemetry, compares it against an expected process model, detects anomalies (chatter, tool wear, thermal drift, feed mismatch, sensor dropout), and gates human-in-the-loop recommendations. A teaching project that makes the monitoring loop concrete.
Open-source, browser-based Ashby material-selection charts. A free alternative to Granta EduPack, built for my Harvard ES-192 course.
Automated crack detection and classification for civil infrastructure (bridges, roads, buildings) using a YOLOv11 object-detection pipeline.
A personalized movie recommender built with Graph Neural Networks, modeling users and items as a single interaction graph.
Remaining-useful-life prediction on NASA's turbofan degradation dataset: time-series modeling for predictive maintenance.
A multi-agent orchestrator-worker-verifier research system on the Anthropic Claude API, built for Boston AI foundry idea-to-product triage: a manager decomposes a question, parallel specialists research it, an independent verifier adversarially checks every finding, and uncertain results escalate to a human, with observability throughout.
A design-to-inspection digital thread for additive manufacturing, built around FFF for my ES 51 students: ingest an STL, choose a build orientation, simulate the build, predict bed-warpage with a thermal-contraction finite-element solve, run DfAM and inspection checks, and emit a gated, machine-readable build record.
Modeling and experimentally validating interface strength in FFF-printed functionally graded materials, comparing direct, interlocked, and gradient transitions.
Doctoral work on low-cost FFF fabrication of multi-material structures with graded properties: design, fabrication, characterization, and multi-scale modeling.
An NSF-funded immersive VR lab built in Unreal Engine for engineering and manufacturing education, with design, fabrication, and testing rooms.
Applied R&D with Flowserve exploring 3D-printed mold patterns as a low-cost alternative to conventional tooling for lost-foam casting.