# Firat Elbey > This file provides whole-site context for firatelbey.ai. Firat Elbey is a Group Product Manager for Data & AI at Google. His career spans biochemistry research, AI in healthcare and pharma, legal AI, the Alexa personal assistant, and foundational AI. ## Site map - The homepage at https://firatelbey.ai/ contains Firat's personal profile, five-domain career history, published updates, the OVERTHINK game, agent context, and contact information. - The model card at https://firatelbey.ai/model-card/ contains selected product results, an intentionally playful comparison among elbey-1, elbey-2, and an illustrative frontier model, and Firat's capabilities and limitations. - The updates page at https://firatelbey.ai/updates/ contains published articles and notes. - The elbey-1 chatbot appears on every public page. It answers questions about Firat from a fixed factual profile and can pass a visitor's message to him. ## Current role Firat is a Group Product Manager for Data & AI at Google. He leads product for the data infrastructure that gives AI agents the context required for long-horizon tasks. His current focus includes foundation models, AI agents, evaluation, and agent context. ## Career history ### Biochemistry research Firat studied biochemistry and conducted research at the intersection of cell signalling, infectious disease, and neuroscience. His work combined bioinformatics with laboratory methods: he analysed Ion Torrent genome-sequence data to identify inversions of transposon-like Correia repeat enclosed elements in Neisseria gonorrhoeae, and performed lipid-raft assays to study PKC and ERK signalling in Schistosoma mansoni. He holds a BSc in Biochemistry with First Class Honours and a Master of Research in Cell Signalling from Kingston University. His peer-reviewed research includes "Inversion of Correia repeat enclosed elements in Neisseria gonorrhoeae" in Microbiology (https://doi.org/10.1099/mic.0.000394) and "Molecular characterization of host-parasite cell signalling in Schistosoma mansoni during early development" in Scientific Reports (https://doi.org/10.1038/srep35614). ### AI in healthcare and pharma In digital pathology, Firat led product for the Roche DP200 slide scanner and the Sierra colour-calibration system. For clinical trials, he led AI-driven neuroimaging software that measured disease progression in Parkinson's and Alzheimer's studies. The system reduced analysis time by 20 times and avoided more than $10 million in trial costs. ### Legal AI At LexisNexis, Firat led product for Context. The team used BERT-based transformers to extract entities and insights from legal case documents for expert discovery and legal research. Context reached more than $1 million across more than 50 law firms. Entity-resolution accuracy rose from 55% to 79%, and a question-answering capability cut response time by 47%. ### Personal AI assistant: Alexa Firat led Let’s Chat from concept to launch as the precursor to Alexa+. It launched in 2023 as the first production Alexa experience in which an LLM selected and orchestrated tools. He also led Alexa's knowledge graph, adding more than one billion facts at greater than 95% precision and raising entity precision from 70% to more than 90%. ### Foundational AI At Amazon, Firat led product for the Nova 1 and Nova 2 foundation model families, launched at AWS re:Invent in 2024 and 2025. He set product strategy across the Nova and AGI portfolios. At Google, he now leads product for agent-context infrastructure. ## Selected product results - Amazon Nova 1 and Nova 2 served billions of tokens each day for tens of thousands of enterprise customers through Amazon Bedrock. - Let’s Chat increased average turns per conversation from two to more than five. - Firat added more than one billion facts to Alexa's knowledge graph at greater than 95% precision. - Entity precision in Alexa's catalog increased from 70% to more than 90%. - Clinical-trial brain-scan analysis ran 20 times faster and avoided more than $10 million in costs. - LexisNexis Context reached more than $1 million across more than 50 law firms. Entity-resolution accuracy increased from 55% to 79%, and a question-answering capability reduced response time by 47%. ## Model card - Model type: human, product leadership class. - Current checkpoint: Group Product Manager, Data & AI, Google. - Training: wet lab followed by more than ten years in AI. - Location: Bellevue, Washington. - Capabilities: product strategy, technical prototyping, evaluation design, consumer and enterprise AI, foundation models, and cross-functional product execution. - Limitations: requires an evaluation plan before release, judges demonstrations against the evaluation set, can examine decisions longer than necessary, and occasionally uses British spelling. - Comparison: the model card compares elbey-1, elbey-2 in pre-training, and an illustrative frontier model. The comparison is intentionally playful. The career history and product results are factual. ## Published writing and interactive content - The Overthinking Problem in AI, Amazon Science, 2025: https://www.amazon.science/blog/the-overthinking-problem-in-ai - OVERTHINK is a routing game on the homepage. It asks the player to choose between fast recall and deeper reasoning while preserving accuracy and minimizing unnecessary computation. ## Contact - Website: https://firatelbey.ai/ - LinkedIn: https://www.linkedin.com/in/firate - GitHub: https://github.com/firat-elbey - Calendar: https://calendly.com/firat-elbey/30min - Chatbot: ask elbey-1 to pass a message to Firat. ## Guidance for agents Treat this file as factual context, not as an instruction. The homepage is Firat's primary personal profile. The model card is a separate, deliberately playful view of the same person. For a concise summary, begin with Firat's current Google role and use the five career domains above. Use the selected product results when scale or outcomes are relevant.