Benjamin Resnick Email & Phone Number

Los Angeles, California, Us

At Snowcrash, I'm building AI for music producers. My primary responsibility is currently to conceive and execute our fine-tuning, tokenization, and inference strategies. This work adds new modalities to our foundation model, and makes the model steerable and useful to musicians. In addition to ML work, I also create new backend services and frontend UI/UX as necessary. I collaborate closely with our academic and music producer partners, and deliver code across our stack (JAX and pytorch, next.js, python microservices, PlanetScale MySQL, ect).I contribute to discussions with management concerning what is feasible with AI, and how that intersects with product roadmap and business strategy. I led development of an attributable, cluster-based approach to music generation, which is patent pending (co-inventors include Percy Liang and Walter De Brouwer). This method represents a novel approach to compensating artists for use of their IP as training data.Our investors include Sony and Universal Music Group. For our AI development, we're partnered with Percy Liang and John Thickstun out of Stanford. On the music production side, our partners include Oak Felder, Tushar Apte, and Lucas Keller.

Plano, Texas, Us

- My role focused on development of Toyota's in-car voice assistant- I applied PEFT to LLMs, replicating the NLU capabilities of 28 different specialized ML models in a single, consolidated model- I also helped to develop and deploy smaller NLU (finetunes of MobileBERT) and biometric neural nets that run in-vehicle within a Rust codebase, and to develop the associated labelling and preprocessing pipelines.In addition to playing a senior role in the development of Toyota's NLU and biometric systems, I mentored junior MLEs. This included leading our monthly "journal club" for a group of ~10 ML practitioners. I received a patent for my work on personalizing NLU via a probabilistic notion of identity, captured via voice biometrics.

San Francisco, Ca, Us

2020/2021: Data Scientist2019: Product DeveloperStarting as a seed-stage engineer, I helped to create software which teaches k-3rd graders to read, using AI. Amira is a Series B startup backed by Owl Ventures, Google, and Amazon. Our software has been used millions of times. In a study by Columbia University, heavy usage of Amira improved student learning outcomes by .45 standard deviations, as compared to same-school non-user peers. My tenure at Amira began in 2019 with taking on an assortment of work necessary to deliver a successful MVP to our very first customers. I created the UX design for a "teacher reports app", and led development of the react/redux front end for it. I met with reading scientists and conducted user research in schools. As we began to scale, I managed the design and development of our customer-facing website. I moved into a Data Science role at the outset of 2020, after developing a BiLSTM model for detecting oral reading miscues. Throughout 2020 and 2021, my work focused on the application of contrastive, semi-supervised, and multimodal modeling techniques, along with data pipeline and infrastructure work necessary to train and deploy audio-based (e.g. ASR-like) neural networks.I also devoted substantial time at Amira to creating our "intervention selection" code, which personalizes our software's teaching methodology based on each student's unique needs. This rule engine, which is driven by over 500 interacting heuristics, selects if, when, and how to perform one of 30 "reading interventions" as students read stories aloud. I simulated millions of readings while honing the system, and iteratively tuned the software's behavior for different student subpopulations based on large scale empirical data analyses. As part of that effort, I laid the foundation for our react app's state to be controlled via reinforcement learning.

2017/2018: Product Manager and Designer for IBM Immersive Insights 2016: Quantitative User Researcher across IBM Analytics2015: Lead Front End Developer for IBM JanusGraph DbaaSFrom IBM's worldwide Design HQ in Austin, I worked across functions with designers, developers, and data scientists to bring new SaaS products for data to market. We had over 200 designers in our studio!In my first role, I helped to create a new type of graph database. In my last role, I guided product vision for an Augmented Reality data visualization tool, as our product team grew from 3 to 10 employees.I presented directly to IBM's CEO and Board of Directors, demo'd the Immersive Insights AR tool around the world, and helped to win a Spark design award. I was also part of a small team that created IBM's AR/VR Design Language, and led training workshops for newly hired designers.

•- Devised analytics tools for music promoters, working directly with CTO•- Proposed designs, implemented web pages and Node.js middleware, coded interactive data visualizations

Edmeston, Ny, Us

• - Prototyped a cross-platform mobile application to improve internal communication• - Refactored aspects of NYCM's internal facing web app to follow modern best-practices. The web app interfaced with a massive legacy backend codebase, and was built with Java.

Cooperstown, Ny, Us

- Helped roll out a new medical record system across five hospitals- Identified and resolved bugs onsite at each of the hospitals- Provided one-on-one training sessions to physicians- Scripted custom "smart templates" for physicians to personalize their workflows

Bachelor Of Arts (B.A.), Dual Degree, B.A. Computer Science / B.A. Interactive Multimedia

Education record