T.J. Trimble

Experience

• Worked with many partners to develop the schema, ontology, and guidelines for tree-based NLU (TOP) from inception through product delivery, enabling the launch of Hey Facebook.
• Developed and maintained model-based and rule-based NLU systems and grammars, improving system accuracy to 95% on complex, multifaceted Q&A domains.
• Developed and maintained ASR language models, grammars, and other components, reducing Word Error Rates on several domains including on search dictation by 50%.
• Wrote and maintained data creation, curation, and annotation guidelines for projects across the stack including ASR, NLU, Q&A, Search, and NLG, used in the collection of millions of rows of data.
• Designed, iterated upon, and shipped new end to end features with my product team such as semantic search, increasing total search-driven revenue by 2% in one quarter.
• Managed a $10 million annual budget delivering quality data for several high impact domains including app search, calling, messaging, reminders, open Q&A, and social across the entire stack, from ASR and NLU to NLG.
• Led a team of Linguists delivering exceptional products on Meta Quest, Meta Portal, and more, owning ASR and NLU components and managing data collections for several search modules.
• Worked with Product Managers, Engineering Managers, and Technology Leads from across the product pipeline (ASR, NLU, Q&A, NLG, etc.) to create project goals, prepare technical plans, and execute.
• Coordinated work with other teams across the product pipeline, from speech recognition (ASR) through language understanding (NLU) to text to speech (TTS), and built internal and external relationships.
• Managed direct reports, project teams, and directed up to one hundred annotators to exceed top level organizational objectives of end-to-end system quality.
• Developed individuals' technical and soft skills through project leadership, goal setting, and mentoring.
• Drove improvements across the product stack through data-driven opportunity analysis and aligning stakeholders around common goals.
• Drove improvements in ASR and NLU development processes across the organization, improving top line quality in some domains from 85% to 95% (-65% error).
• Triaged high visibility bug traffic through data-driven blame processes to get alignment on priorities.
• Created and developed hiring goals, job postings, core competencies, and interview questions.
• Conducted dozens of interviews, additionally acting as hiring coordinator for dozens of contractors.
• Developed technical and non-technical roles in collaboration with key stakeholders and peers.

• Developed CLI and web-based tools for ontology development, annotation quality, data quality, annotator training, inter-annotator agreement and model metrics, guidelines management, and annotation visualization in Python, PHP/Hack, and JavaScript/React.
• Worked with highly cross-functional team to implement a zero-to-one voice app management system with complex UX design, storage budget, and system design constraints, supporting the refresh of wit.ai
• Designed and implemented annotation tools for collection and annotation tasks for NLP in React and Hack used to collect millions of annotations by hundreds of annotators across dozens of projects.
• Led a four person engineering team, leading design discussions, organizing projects and workloads, and providing technical leadership and mentorship.
• Coordinated with other teams, built internal and external relationships, and developed individuals' skills.
• Consulted on contracts for Quality Assurance and technical requirements.
• Created and developed hiring goals, job postings, core competencies, and interview questions for technical and non-technical roles in collaboration with key stake holders and peers.
• Conducted and reviewed dozens of interviews.

• Worked closely with client to develop schema, ontology, and guidelines for tree-based, task oriented semantic annotation (TOP) from model inception through product delivery of Meta Portal.
• Annotated open class Q&A, navigation, and communications domain data with dependency parses and TOP labels.
• Developed tools to improve annotation quality and speed, including a custom annotation UI and annotation validators.

• Owned and improved an Information Extraction engine processing thousands of documents every day, built on top of Stanford's TokensRegex system.
• Designed, developed, and deployed a system matching named entities with topics over 100,000 documents per day.