Firat Elbey

elbey-1 / model card

Model card

This model card summarizes selected results, comparative attributes, capabilities, and limitations from more than a decade of building AI products.

Model type

Human, product leadership class

Checkpoint

Group PM, Data & AI, Google

Training

Wet lab, then 10+ years in AI

Location

Bellevue, Washington

01 / Results

Metric highlights

The figures below show the scale and performance of the products I led.

Billions

Tokens served each day by Amazon Nova 1 and Nova 2The daily volume was comparable to the full text of English Wikipedia.

Tens of thousands

Enterprise customers using Nova through Amazon BedrockThey accessed Nova through Amazon Bedrock.

2 → 5+

Average turns per Let’s Chat conversationLet’s Chat increased the average conversation from two turns to more than five.

1B+

Facts added to Alexa's knowledge graph at 95%+ precisionAt one fact per second, the same work would take 30 years.

70 → 90%+

Entity precision in Alexa's catalogThe error rate fell by more than two-thirds.

$10M+

Costs avoided in Parkinson's and Alzheimer's drug trialsBrain-scan analysis ran 20 times faster.

02 / HUMAN VS FRONTIER

Model card

The human column records my experience. elbey-2 is still in pre-training, and the frontier model column is illustrative.

Comparison of elbey-1, elbey-2, and a frontier model across training, knowledge, product deployment, judgment, embodiment, and safety.
Open the full-size image to zoom.

† elbey-2 remains in pre-training. No benchmark results are available yet.

‡ The human input count is an order-of-magnitude estimate. Exact model training counts are not public.

The comparison is intentionally playful. The career history and product results are factual.

03 / Capabilities and limitations

Capabilities and limitations

I combine product strategy with technical prototyping and require an explicit evaluation plan before release.

Capabilities

  • I lead AI products from initial definition through adoption by consumers and enterprises.
  • I have authored more than 10 PR/FAQs and business cases reviewed at the C-suite level.
  • I set product strategy across organizations led by several senior vice presidents.
  • Within six weeks of joining Google, I built a multi-agent application.

Limitations

  • I will not release a product without an evaluation plan.
  • I judge demonstrations against the evaluation set rather than the strongest example.
  • I tend to examine decisions longer than necessary.
  • I occasionally use British spelling.