Designing and implementing custom data acquisition platform, supporting intermittent connection and live data analysis.Designing and implementing traction control system for four-motor 4WD vehicle.

Built launch control system for rear-wheel driven Formula Student Electric vehicle.- Implemented PI-based controller for wheel slip ratio-based feedback control for the FSAE acceleration event- Designed and implemented custom vehicle simulations for control algorithm design and tuning- Implemented data analysis from recorded vehicle data for tuning

MIT Sandbox Innovation Fund (Spring 2024 Cohort)

Working with NVLink architecture team on functional coverage support for testing of the high-speed GPU-to-GPU memory fabric.- Designing and implementing automatic coverage code generation, using common bus spec to minimize additional code maintenance.- Adapting existing specification tooling to fit seamlessly into the existing coverage workflow.- Designing and implementing architectural test in order to validate and improve results from coverage reporting.- Configuring relevant coverage points of interest for specific NVLink revision, using implementation spec.

Held approximately eight office hours a week, helping teach class content and labs for 6.1910, a core Computer Science class at MIT.The labs start with combinational and sequential circuits, and culminate with RISC-V single-cycle and pipelined processor implementations. This includes hardware for memory caching and software for a simple operating system.

Worked as an electrical and reliability engineer for consumer-focused phased array antenna products- Electrical verification of DC power electronics- Automated functional tests for reliability testing during product development- Ground up PCB design including ethernet routing with PoE and impedance matching- Coordinated with suppliers and manufacturers for PCB assembly- Worked with reliability engineers to design configurations to their specifications

Undergraduate Researcher for the CSAIL Spoken Language Systems, working under Hongyin Luo and James Glass.Built and co-designed an Open-Domain Question Answer system, comprised of a separate answering API server and interaction web server. Answers are predicted through a multi-step process, including query expansion, document retrieval, reranking, and reading. Each request supports live customization of components in the QA pipeline. Any text database can be used, such as Wikipedia or Investopedia, once a corresponding BM25 index is pre-generated.The systems below were optimized to run on a server with two Nvidia Titan XP (12GB) GPUs, for ~10s extractive answering from Wikipedia:- GAR query expansion- BM25 passage retrieval- BERT-based passage reranking- LLM or ELECTRA-based passage reading to produce an answerContributed to review of research paper: “Interpretable Unified Language Checking”, which focuses on prompting LLMs to evaluate fairness and fact-checking of statements.

During fall 2021 to summer of 2022, I worked as a full-time Software Engineer at Plumerai, the microcontroller-focused machine learning startup, during my gap year in London.I co-designed the new people detection ML data warehousing for the team’s new training requirements, ultimately built on Snowflake and Google Cloud Storage. The new methods to ingest and export data increased efficiency and flexibility for training data selection and analysis during the researchers’ development processes.Meeting with representatives from external service providers, combined with internal benchmarking, allowed me to evaluate and present the opportunities offered by such applications. Through this process, I led the team’s key migration from Postgres to Snowflake. Following Plumerai’s standard procedure, I had regular meetings with Plumerai’s ML researchers to update product specific targets. Strict code review requirements allowed me to both provide and apply feedback on new features and fixes across the company.Working with the marketing team, I deployed both static webpages and outreach-focused applications from scratch One notable example is the ‘people detection’ landing page- one of the core product focuses of Plumerai in 2022. Additionally, I built a fully automated public benchmarker, providing results from Plumerai’s proprietary Larq Compute Engine for Microcontrollers. I worked closely with external designers and applied internal feedback to deliver a polished product.Importantly, I focused on maintaining secure infrastructure in order to limit unintentional and irreversible operations. It was crucial to be aware of resource consumption and data, and the infrastructure therefore had multiple layers of protection against unintended operations. Credential privileges were provided on an as-needed basis, further limiting the potential impact of rogue systems. All cloud data was frequently backed up and validated, to both catch issues and recover quickly.

Co-founder and main developer of Candivote (candivote.org). Currently developing full user interface in React, utilizing Firebase to handle data, authentication, and hosting operation. Working with three Undergraduates from Brown and Yale University. Focus on dynamic, functional, and appealing page layouts.

Hired as a contract Python architect and engineer. Contributed to development of DGNet Control, aimed to facilitate remote management of university lecture halls. Collaborated on multiple systems including, architecting a Raspberry Pi power management module and a password-less QR code web authentication, using strict architecture development and review.Pitched and built next-generation execution stack tracking, preventing program halts due to exceptions. Built to be backwards compatible with existing system, while still adding its valuable functionality. Now actively used in the new codebases within the company.

Wrote new guides for current and future employees to properly construct documentation and style the final code. Built a Chrome plugin facilitating the display of webcam feeds in browser-based presentations. Worked to design the architecture and code for an interface for Firestore Native, in which the other developers had no experience.