Emmanouil (Manos) Koukoumidis

CEO @ Oumi - www.oumi.ai On a mission to empower every organization to take their AI future into their hands; build and compound their own specialized intelligence that they can fully own and control.

- Bootstrapped and launched Large Language Models for Cloud setting up and leading efforts across ~30 teams across numerous Google Product Areas (divisions) including almost the full spectrum of GenAI development (SFT, RLHF, RAI, serving infra, parameter efficient tuning, APIs, documentation, pricing and others) - Generative Language API (https://blog.google/technology/ai/ai-developers-google-cloud-workspace/) - Cloud Natural Language: Democratizing AI with Google Cloud Natural Language APIs, Google Cloud Natural Language AutoML, Translation API, AutoML Translation. - Document AI Modeling: Automating enterprise document workflows by using state-of-the art Computer Vision and NLP to extract structured data from unstructured documents (Google Cloud Document AI). Proposed and led DocAI Workbench (Custom Document AI Processors) from inception to launch.

Engineering and Science for Biscuit Labs (pre-Series A smart home startup in stealth mode) Biscuit Labs Homepage: http://biscuit.io/

Portal Assistant Team, AR/VR Group - Bootstrapped and spearheaded Portal Assistant’s integrity efforts. Drove successful exec review receiving up to 45 engineer headcount. Established integrity review processes across the Portal Assistant organization. - Defined the vision for assistant “integrity” with leadership team aimed at protecting user trust in Portal. Led cross-team efforts and established collaborations to build integrity solutions across several features and products in the AR/VR and Messenger organizations. These products included Portal ASR/NLU/NLG/TTS, Messenger Proactive Intelligent Suggestions, Portal Real-Time Communications, VR Social Experiences. Integrity solutions leveraged both ML- and rule-based NLP techniques operating both proactively and reactively to ensure that only safe content is shared across users or between users and the digital assistant. - Bootstrapped Portal assistant “Personality” efforts in collaboration with PM, Design, and Content Strategy.

Led teams of engineers and designers to deliver innovative machine learning-based services and products. Led the engineering operations (hiring, agile processes, CI/CD) and science for Zo (www.zo.ai), an industry-leading open-ended multimodal chatbot (text, voice, images). Zo.ai was the "ChatGPT of 2015" using retrieval and generation to do multi-turn and multi-modal conversations on any topic. Pioneered artificial emotional intelligence technologies and services across Microsoft. Bootstrapped effort and built team with executive funding. Architected a scalable ETL pipeline for person entity conflation on Azure for undisclosed partner using Hadoop (MapReduce, HBase, Hive). Built the ML-based entity matching algorithm. Achieved ~3x speedup, +9% accuracy, +16% recall. Advisor in partnership with undisclosed vehicle manufacturer on intelligent driver-assistance services.

Led the development of the location-based reminders (Geofencing) for Cortana and other intelligent location-based services. Invented and architected the dynamic geofence capability (featured in tech news as key differentiator). Bootstrapped and led the search-along-a-route V1 project starting it as hackathon project and shipping it with the "Via" Windows Phone app.

Advisor for Playtabase that is developing the Reemo wearable.

Developed innovative driver-assistance services in collaboration with MIT, NUS, SMART and Singapore's transport authorities. Received awards. Broader impact: The developed system demonstrated that compute-intensive intelligence that was once thought to belong to the cloud, could actually be performed on smartphones with several advantages (lower resource consumption and infra cost, improved privacy).

Thesis: Collaborative and adaptive mobile device-resident service architectures Intelligent edge & intelligent cloud service architectures for 1) improved resource efficiency (CPU, memory, storage, energy), 2) lower response time, 3) lower infrastructure and data center operating costs, and 4) improved user privacy. This research seeded a lot of the ideas for what was seven years later called "edge intelligence." For more information see the "Pocket Cloudlets" and "SignalGuru" papers that demonstrate how memory/storage-intensive and compute-intensive intelligent applications could be supported on smartphones (Windows Mobile, iPhone 3).

Thesis: Development of an extensible network platform for real time data collection and fusion in intelligent vehicles' networks