Bounded autonomy
Security agents should act only within explicit authority, scope, and control boundaries.
AI × Security Engineering
Making autonomous security validation trustworthy.
I build and lead agentic AI systems for bounded, evidence-driven security testing—helping teams move from vulnerability discovery to verified exploitable risk.
Authorized by design
Verified by evidence
Explainable in operation
Technical thesis
Trust is a system property.
Autonomous security testing needs more than capable models. It needs enforceable boundaries, defensible evidence, and operator control.
Verified exploitability
Findings become useful when evidence distinguishes theoretical exposure from demonstrated risk.
Agentic system security
Identity, authorization, tool use, memory, and human intervention belong in the agent threat model.
Operational trust
Reproducibility, audit trails, and explainable decisions turn autonomous testing into accountable practice.
Selected product work
Built systems, documented reception.
These case studies separate my contribution from external evidence about the product, selection, or use.
Multimodal AI · Communication coaching · 2019–2023
TalkMeUp
Co-founder and CTO
An AI-powered communication-coaching platform that turns video, audio, and language signals into structured feedback. The product was designed to make repeated practice and individualized guidance available beyond a live classroom or coach.
What I contributed
- Led product and engineering work from early concept through a scalable, cloud-delivered system.
- Designed the multimodal analysis workflow across video, audio, transcripts, and derived communication signals.
- Integrated deep-learning models, language models, signal-processing models, and rule-based engines into one user-facing product.
- Modernized the platform from a monolithic application into a cloud-native distributed-services architecture.
- Established automated delivery and quality-control foundations for repeatable product releases.
- Worked across model development, backend architecture, product delivery, engineering leadership, and enterprise deployment rather than treating AI as an isolated prototype.
What public sources validate
- Public patent records name Yi as an inventor on two granted U.S. patents for interactive AI analytical systems.
- Co-author of an ACM ICMI 2020 paper on multimodal enthusiasm coaching.
- TalkMeUp received three Gold 2022 Brandon Hall technology awards, including Best Advance in AI for Business Impact.
- Training Industry selected TalkMeUp for its 2023 Top Advanced Learning Technology Companies list.
- SXSW selected TalkMeUp as a 2021 Future of Work Pitch finalist.
- ASU+GSV selected TalkMeUp among the 2021 GSV Cup Elite 200 semifinalists.
- Wellesley, St. Lawrence, and TCU faculty publicly described using the platform with students.
- A Hearst executive publicly described selecting and expanding use of TalkMeUp for leadership communication.
- The product is publicly distributed through Zoom App Marketplace.
- The product is also publicly listed in AWS Marketplace.
- TalkMeUp's 2018 UpPrize win and $150,000 grant predate Yi's documented 2019–2023 tenure.
Selections, customer quotations, and marketplace availability document product-level reception. Yi’s contribution is stated separately; none of these records is presented as a personal award or as proof of individual influence by itself.
Applied AI · Children’s nutrition · 2018–2019
LittleMoochi
Co-founder and engineering contributor
A free, mobile-first learning experience that used an AI-powered virtual pet to make healthier eating choices concrete and engaging for children. Meal photos became the input to a loop connecting food recognition, nutritional guidance, and play.
What I contributed
- Helped build the cross-platform mobile application and cloud-native backend from the ground up.
- Led development of multimodal neural-network components for recognizing food ingredients from meal photos.
- Connected recognition results to a personalized food-recommendation system and the virtual-pet interaction loop.
- Worked across the application, backend, data model, cloud infrastructure, and machine-learning layers of the product.
- Supported partnerships, user outreach, and community programs in addition to the engineering work.
- Participated with the team in NSF I-Corps and the CMU/Ascender entrepreneurship ecosystem, using external feedback to refine the product and its delivery model.
What public sources validate
- Carnegie Mellon lists Yi on the 2019 NSF I-Corps LittleMoochi team.
- CMU records LittleMoochi as the 2019 McGinnis Venture Competition graduate-track runner-up and names Yi on the team.
- Forté reports that LittleMoochi won both judges’ selection and audience favorite awards in its 2019 Power Pitch.
- Rice selected LittleMoochi for its 42-team 2020 Business Plan Competition field.
- Ascender reports incubator participation, TiE Pittsburgh finalist selection, and a third-place LiftOff PGH result with a $10,000 prize.
- CASTP's retrospective records a first-place result in its 2019 technology competition.
- CAST-USA awarded LittleMoochi its 2023 Outstanding Innovative Enterprise Award and identifies Yi Xu as the company representative.
- Food Helpers and independent local reporting document continued school use of the app years after the founding period.
- The LittleMoochi app remains publicly listed in Apple's App Store.
These are team, product, or program records. School, student, user, and snack-pack counts vary by source and date, so this page does not merge them or convert them into Yi’s personal impact metrics.
Recognition and reach
Recognition earned. Impact documented.
Awards matter most when the record also shows what was built, who used it, and how the evidence connects to my work. Every item below carries its attribution and timing.
Evidence-led impact
The latest reach report is shown once. Historical reach milestones remain in the evidence record, while distinct outcome, program-scale, adoption, and distribution measures stay separate.
78,000
students reached
A program stakeholder directly thanked Yi for his contribution to LittleMoochi’s growth and reported this continuing reach milestone.
Preserved private correspondence; the measurement definition and total are not yet independently audited.
Source preserved privately75,000+
school snack packs
Food Helpers documented the scale of the school program in which the app supports nutrition education.
Snack packs are program activity—not app users or Yi’s personal reach.
View source77% / 63%
reported behavior outcomes
The letter reported that 77% of sustained users maintained or improved weekly nutrition scores, and more than 63% of recommended foods appeared in intake within two weeks.
Letter-reported results; underlying data and methodology have not been independently audited.
Source preserved privately3
universities
Faculty from Wellesley College, St. Lawrence University, and Texas Christian University described using TalkMeUp with students.
Institutional use is separated from company-reported performance multipliers.
View sourceHearst
enterprise adoption
A Hearst executive publicly described selecting and expanding TalkMeUp for leadership communication development.
Customer use is documented; performance figures remain company-reported.
View source2
public marketplaces
Public listings made the platform discoverable and deployable through established cloud and collaboration ecosystems.
Marketplace availability is distribution evidence, not sales or endorsement.
View sourceComplete recognition record
Awards, finalist placements, and competitive selections
Product and team recognition is not presented as a personal award. The labels show whether an item occurred during, before, or after my documented role.
TalkMeUp
Yi served as co-founder and CTO during 2019–2021 and 2022–2023.
- 2021SXSW Pitch — Future of Work finalistExternal selection · during tenure
- 2021GSV Cup Elite 200 semifinalistExternal selection · during tenure
- 2022Three Brandon Hall Group Gold awardsProduct awards · during tenureBest Advance in Unique Learning Technology · Best Advance in Sales Enablement and Performance Tools · Best Advance in AI for Business Impact
- 2023Top Advanced Learning Technology CompanyTraining Industry selection · during tenure
- 2018UpPrize Great Solutions Challenge — first place and $150,000 grantCompany history · before Yi’s documented tenure
- 2024Top AI in Training Companies selectionCompany recognition · after Yi’s documented tenure
- 2024EdTech Awards — artificial-intelligence solution finalistCompany recognition · after Yi’s documented tenure
LittleMoochi
Yi co-founded the product and led core app, backend, and AI development.
- 2019NSF I-Corps site-team selection at Carnegie MellonInstitutional selection · Yi named on team
- 2019McGinnis Venture Competition — second place, graduate trackTeam award · Yi named on team
- 2019Forté Power Pitch — judges’ selection and audience favoriteTeam awards · $10,000 total
- 2019CASTP technology competition — first placeTeam award · organizer-verified
- 2020Rice Business Plan Competition — selected among 42 teamsExternal selection
- 2020Ascender, TiE Pittsburgh, and LiftOff PGHIncubator · finalist · third place and $10,000Ascender incubator participant · TiE Pittsburgh Startup Showcase finalist · LiftOff PGH Ideathon third place
- 2023CAST-USA Outstanding Innovative Enterprise AwardEnterprise award · Yi represented LittleMoochi
Evidence standard: public organizer, institutional, customer, marketplace, or publisher records where available. The July 2026 LittleMoochi reach figure is stakeholder-reported from private correspondence and is not yet independently audited.
Public work
Security writing, research, and invention.
Public records with stable source links and precise authorship or inventor attribution.
Defending the Next Frontier of AI Infrastructure Threats — MCP Security
A public analysis of the attack surface created when AI agents invoke tools and connect to sensitive systems, published by Ridge Security under the Oliver Xu byline.
On Leveraging the Visual Modality for Neural Machine Translation
One of four equal-contribution authors of published research on visual context in neural machine translation using the How2 dataset.
On Leveraging Visual Modality for Speech Recognition Error Correction
Equal-contribution co-author of a How2/ICML workshop paper on visually grounded correction of automatic speech-recognition errors.
Spark Creativity by Speaking Enthusiastically
Co-authored multimodal research on AI-assisted feedback for communication coaching.
Building JDMatcher: LLMs and Workflow Engine for Job Seekers
A public proof of concept and implementation note combining workflow orchestration, document parsing, web extraction, and LLM-assisted job-description analysis.
AI Evolution and Foundation Models
A recorded Silicon Valley Entrepreneurship Forum talk on AI evolution and the role of foundation models in progress toward more general systems.
Inventive record
Interactive AI systems
Named inventor on patent families for AI systems that analyze multimodal communication and generate structured feedback.
About
I translate ambitious AI ideas into systems people can operate, inspect, and trust.
Formal name: Yi Xu · Professionally known as Oliver Xu
I’m an AI and software engineering leader with more than two decades of experience building products and leading teams across cybersecurity, AI infrastructure, communication intelligence, and enterprise software.
My current focus is trustworthy agentic AI for autonomous penetration testing and continuous security validation. Earlier work spans large-scale infrastructure optimization, machine-learning retrieval, multimodal analysis, and AI-powered communication coaching.
Carnegie Mellon University
Master of Information Systems Management
University of Science and Technology of China
Master’s and Bachelor’s in Computer Science
Career throughline
From enterprise systems to trustworthy agents.
2025–present
Ridge Security Technology Inc.
Senior Director of Engineering
Agentic AI security and autonomous security validation
2024–2025
Meta
Core System Engineering Manager
Large-scale infrastructure planning and optimization
2019–2023
TalkMeUp
Co-founder and CTO
Multimodal AI communication assessment and coaching
2021–2022
ElectrifAi
Senior Director
Machine-learning products and backend engineering
2018–2019
LittleMoochi
Co-founder
Applied AI for children’s nutrition education
2017–2018
PwC
Senior Manager, AI Solutions
Enterprise machine-learning solutions and team building
2005–2017
Hewlett Packard Enterprise
Engineering manager and software engineer
Enterprise discovery software and globally distributed engineering
Connect