Luke Hornof
About
Detail
AI Engineering Leader | Founder and VP of Engineering | Intel (Nervana) | Ph.D. in Computer Science
California, United States
Résumé
Jobs
verified_user
0% verified
- CAI Engineering & LeadershipConsultingJul 2023 - Current (3 years)• Provided business and technical expertise to AI community, companies, and investors. • Delivered invited talks at events such as the AI Product Creators AI Fusion Summit. • Offered subject matter expertise for investment firms including DE Shaw and GLG.
CEO and co-founderLuminideJan 2020 - Jan 2022 (2 years 1 month)Acquired by Akridata in 2022. Luminide makes it easier to improve the AI models that are being used to solve some of the world's biggest challenges, e.g. models used for cancer radiation therapy.- Senior Software EngineerBandPage acquired by YouTubeFeb 2012 - Jul 2013 (1 year 6 months)Built REST and graph APIs, artist search, and messaging infrastructure supporting 500,000 artists and 30 million monthly active users. (Java, Spring, MySQL)
- Senior Software EngineerAdBriteNov 2009 - Nov 2011 (2 years 1 month)Built core ad serving components — inline, XML, and real-time video ads — for one of the largest independent ad exchanges, serving 150 million impressions daily across 60,000 sites.
- Senior Software EngineerSun MicrosystemsApr 2004 - Nov 2009 (5 years 8 months)Built software for an advanced video server capable of storing 300,000 hours of content and delivering 160,000 simultaneous 2Mbps streams. (C++, Linux, TCP/UDP)
- KSenior Software EngineerKealia IncFeb 2003 - Apr 2004 (1 year 3 months)Built software for next-generation digital TV at Andy Bechtolsheim and David Cheriton's startup. Acquired by Sun Microsystems in 2004.
- Senior Software EngineerTransmetaJul 2000 - Jul 2002 (2 years 1 month)Implemented a C interpreter for the AMD-64 instruction set, requiring deep hardware knowledge and delivered on an aggressive timeline. (C, Java, x86)
- Research ScientistUniversity of PennsylvaniaFeb 1998 - Jul 2000 (2 years 6 months)Developed self-specializing mobile code that automatically optimizes itself at its destination — extending my Ph.D. research in program specialization with a novel type system enabling formal safety guarantees at runtime.
- Chief Technology OfficerThirdLaw Runtime AI SafetyThirdLaw is building runtime AI safety for the enterprise — giving IT and security teams visibility and control over LLM and agent behavior. I led the engineering organization, built out the team, and drove development of the core platform.
- Founder & CEOLuminideLuminide was a cloud-based IDE for AI model development — faster than Colab, easier than SageMaker, with built-in experiment tracking, hyperparameter tuning, and visualization. We bootstrapped, built the product, and drove go-to-market through community engagement on Kaggle, fast.ai, and social media — reaching 100 signups in 6 weeks and 143K tweet impressions at launch. Selected as a finalist in the Mercury Raise First Check competition. Acquired by Akridata in 2022.
- Director of Software, AI ProductsintelAfter the acquisition, I scaled the Nervana software team from 10 to 150 engineers across the US, Poland, India, and Israel. Building a team that fast required a sophisticated hiring process — screening thousands of candidates, running coding challenges, and where possible acquiring intact teams rather than hiring one by one. Culture across four countries was harder. I discovered that "technical fights" were really trust issues between engineers who had never met. The fix was simple: make them travel. Friendships formed, mutual respect followed, and the fights disappeared.
- Senior Director, AI ProductsAkridataFollowing Akridata's acquisition of Luminide, I led the integration of Luminide's technology into Akridata's platform — owning product, go-to-market, and community engagement. Drove a 50x increase in social media engagement and built community presence across Kaggle and ML Commons.
- STechnical ConsultingSoftware Engineering and Artificial IntelligenceAdvised companies and investors on software engineering and AI. Gave invited talks including the AI Product Creators' Fusion Summit and provided technical analysis to expert networks including DE Shaw and GLG.
- Director of AI SoftwareNervanaNervana was the world's first AI hardware startup — founded on the belief that a chip designed from scratch for AI could outperform repurposed graphics hardware. I founded the software team, personally built the initial interpreter and driver for our custom chip, and grew the team to 10 engineers before Intel acquired us in 2016. Our work supported novel hardware innovations like Flexpoint, a numerical format designed specifically for deep learning and later published at NeurIPS.
- VP of EngineeringTensorWaveLed engineering team, recruiting and technical efforts, datacenter and SAS operations.
Education
verified_user
0% verified
- Ph.D, Computer ScienceUniversité de Rennes IJan 1993 - Jan 1998 (5 years 1 month)Thesis: Static Analyses for the Effective Specialization of Realistic Applications (Advisor: Charles Consel)
- B.S, Computer Engineering, Computer ScienceCarnegie Mellon UniversityJan 1988 - Jan 1992 (4 years 1 month)Senior Honors Thesis: Compiling Prolog to Standard ML (Advisors: Peter Lee and Frank Pfenning)
Publications
verified_user
0% verified
- UUS Patent 10,552,119 — Dynamic management of numerical representation in a distributed matrix processor architectureUnited States Patent and Trademark OfficeApr 2020Patent covering dynamic numerical representation management in AI hardware, building on the Flexpoint research developed at Nervana/Intel.
- AFlexpoint: An adaptive numerical format for efficient training of deep neural networksAdvances in Neural Information Processing Systems NIPSDec 2017Introduced Flexpoint, a novel numerical format designed specifically for deep learning that outperforms traditional floating point on AI hardware. Contributed as part of the Nervana engineering team. 388 citations.
- PA uniform approach for compile-time and run-time specializationPartial Evaluation International Seminar Dagstuhl CastleFeb 2005Presented a unified framework for program specialization applicable at both compile time and runtime. 191 citations.
- HCertifying compilation and run-time code generationHigherOrder and Symbolic ComputationDec 1999Developed type-theoretic foundations for safe runtime code generation, enabling self-specializing mobile code to optimize itself at its destination. Co-authored with Trevor Jim. 69 citations.
- PA Study of Large Object SpacesProceedings of the ACM International Symposium on Memory Management ISMMOct 1998
- ATempo: Specializing systems applications and beyondACM Computing SurveysJan 1998Presented Tempo, a specialization system for C programs that demonstrated practical performance gains across real-world systems applications. 155 citations.