Implemented a Next.js system on the T3 stack (tRPC, Tailwind,
Prisma, PostgreSQL) with TOTP-encoded dynamic Data-Matrix barcodes
Features offline caching for weak-network environments and Google
OAuth; deployed serverlessly on Vercel for seamless use by
non-technical staff
Successfully tested in a 120-student underground auditorium during
BU CS391, leading the professor to adopt it for next semester and
invite me as a course assistant
Interactive CPU Learning Simulator
Built an interactive teaching CPU (DEC Beta ISA, a simplified RISC
architecture from MIT) with step execution, live register/memory
editing, and animated datapath visualizations
Developed using React for browser-based interactive learning
experience
Released as open-source web/desktop app; adopted in BU's CS210 and
part of the MIT teaching tradition via OpenCourseWare
Optimizing BU Professors' Classroom Allocation to Minimize Walking
Cost
Built a Python/Numpy simulated-annealing scheduler with
campus-scale distance modeling using OSMnx
Achieved 97% reduction in faculty walking distance across five
weekdays through optimization algorithms
Processed and cleaned 200k+ SIS course and classroom records into
optimization-ready datasets in under 20 minutes
As president, directed development of a full-stack student
platform: event registration, sponsorship management, real-time
talent-show voting (handling 500 concurrent users and 2k+
websocket requests/sec), accounting tools, and a secure chair
election portal.
Built an LLM onboarding assistant by engaging the student
community to contribute multilingual QA pairs and fine-tuning
Qwen-2.5 with RAG on BU resources, creating a collaborative
onboarding tool for FAQs on housing, transport, culture, and
careers.
Platform serves 1k registered users, 3.5k unique monthly visitors,
and 77k monthly requests; enabled cross-department collaboration
within a 120-member association, partnerships with campus clubs,
BU administration, and the Eastern/Central US CSSA Alliance.
Recruited new developers, mentored junior members, and drove a
cloud cost-reduction strategy—cutting spend by 30% through
diversifying infrastructure and reducing reliance on AWS.
RFID wristband lap counting system for running
Using C# drivers to communicate high-frequency RFID antennas,
processing data with Python, and exporting to Excel.
Reducing congestion at the starting point of the circular track by
at least 50% and compressing the daily 20-minute data analysis to
an instant task