Los Angeles, California, United States
Currently developing a meta-harness for transforming traditional organizations to AI-Native, empowering them with the rapidly expanding frontier of AI capability, compounding knowledge and process from an individual to an institutional level.
Building a meta-harness to support Agentic AI transformations at an organizational level across all business functions, primarily a system for optimizing process and compounding institutional knowledge by maximally utilizing frontier capabilities in a completely vendor-agnostic and execution-agnostic methodology -- filling gaps in capability as they exist in any particular moment, while remaining flexible to the rapidly increasing capabilities of frontier offerings.
- Developed knowledgebase extraction technologies for large unstructured documents, including 400-page PDF documents detailing health insurance policy details - Optimized myriad agentic functions within the product, identifying key gaps between historical capability of frontier offerings at time of implementation vs. present state - Proposed a future-proofed framework of vendor-agnostic primitives for describing agentic capability with the flexibility to adapt to the rapidly changing AI ecosystem
- Performed research on state-of-the-art generative AI techniques, models, and tooling, and assessed for suitability to be incorporated into main-line products - Developed PoCs using generative AI to test and demonstrate feasibility, and implemented viable projects in production, such as a domain-agnostic aspect-based sentiment analysis for Amazon reviews using LLMs - Developed a framework to evaluate the impact of new AI technologies on existing business functions to make recommendations for defensibility and identify opportunities for optimization
- Implemented a Snowflake data ingestion engine, which pulls data from multiple heterogeneous organizations within the Pacvue portfolio (via an S3-based data lake). - Implemented a data modelling and transformation layer based on DBT and Dagster that models the raw ingested data into a well-documented and consistent structure. - Implemented an ML-Ops program, based on MLFlow and Dagster, that enables data scientists to effectively develop, iterate, productionize, and maintain machine learning models. - Developed and productionized a novel Sales Estimation Model for predicting the sales volumes of products sold on Amazon, which is used in several core products across Pacvue assembly companies. - Advise engineering teams on architectural and engineering best-practices, and mentor promising young engineers in their personal development. - Advise product management teams on technical feasibility, scoping, and program management.
- Implemented a data ingestion engine for loading data from production databases into Snowflake. - Advised on architectural decisions for a diverse set of technical teams across several Helium 10 products. - Designed a novel declarative data orchestration engine, and developed a working proof-of-concept that was capable of ingesting and transforming several terabytes per hour of raw data in a flexible set of input formats and storage engines, and delivering it to external destinations efficiently.
- Upon acquisition, trained new technical leaders on existing software architecture, and advised them on architectural best practices. - Advised executive team on product and business decisions. - Assisted in responses to production emergencies as needed.
- Originally hired as CTO and sole-developer for AddressReport (originally Rentenna), a web app that aggregated real estate listings and public data to provide detailed insights to empower consumers to make informed real estate decisions, especially in the New York City real estate market. This company was eventually acquired by Onboard Informatics. - Performed a full rewrite and modernization (originally as sole developer) of FranchiseHelp, a web application that enables prospective entrepreneurs to research franchising opportunities and connect with those brands. - Developed from scratch Union Square Linens, which was a bulk wedding linens retailer. As part of this project, developed a machine vision technology that enabled the projection of different fabric colors and textures onto photographs, which dramatically reduced the cost of accurately representing visually hundreds of thousands of combinations of fabrics, shapes, and sizes of linens. - Developed from scratch Small Business Post, which was a web portal enabling small business owners to list their businesses for sale to new owners. - Acquired two existing web apps, Dandelion and YC-backed Boostable, that enabled small retailers on platforms such as Shopify, Etsy, and Eventbrite to advertise on social media platforms. - Rehabilitated a damaged email sending domain reputation and developed a sophisticated email orchestration engine that was eventually able to send upwards of one million emails per day. - Hired, organized, and directed an engineering team of 8 engineers at its peak. - Developed product roadmaps in concert with business teams. - Engaged in pitch meetings as technical expert with partners, customers, investors, acquisition targets, and auditors. - Replaced light-bulbs, unclogged drains, ensured team members did not leave their dishes in the sink
- As a fresh college graduate, was given the opportunity to re-architect Workday's proprietary multi-tenant internal search engine from the ground up to address major performance and correctness shortcomings (according to an insider source, this was still being used as late as 2017, shockingly!) - Conceived, designed, and tested a general-purpose mechanism for observing and propagating data changes through a complex proprietary in-memory graph database that enabled for the search engine index to be updated in real-time when search objects changed. - Developed a framework for assessing sensitive information leaks that could occur when searching over data subject to heterogeneous data access policies.