How do you protect personal data while keeping AI training data useful?
Airbnb built a system that detects and replaces sensitive information at scale, without losing the context models need to learn.
Here’s how it works. 👇
airbnb.tech/infrastructure/d…
1
5
1,839
A forecast is only the beginning. What happens when numbers meet the real world?
Harrison from Airbnb’s Data Science team shares how we approach forecasting, turn models into business decisions, and put data to work in a Q&A with Significance magazine.
academic.oup.com/jrssig/arti…
Skipper is now open source! 🎉
Explore Airbnb’s embedded workflow engine, now available to developers under the Apache 2.0 license on GitHub.
medium.com/airbnb-engineerin…
What if you could get frontier-level AI classification at 1% of the cost?
Airbnb engineers fine-tuned a small language model to deliver comparable accuracy in under 100ms. Here’s how. 👇
airbnb.tech/ai-ml/frontier-q…
Airbnb Engineering retweeted
Everyone’s excited about decision models. When should you buy one, or train your own?
We fine-tuned a 2B model for a production classifier: frontier-level quality, 35ms, ~1% of the cost.
We’re sharing when we think it’s worth it, and when it isn’t: airbnb.tech/ai-ml/frontier-q…
How do you safely launch an AI support agent globally? 👀
Airbnb engineer @WeipingPeng shares the story at #QConSF on Nov 16th, from guardrails to outsmarting attackers to lessons that shaped every agent launch.
qconsf.com/presentation/nov2…
4
1,944
How do you give AI agents institutional knowledge?
@Airbnb CTO @Ahmad_Al_Dahle shares Project Everest, a “company brain” built from employee workflows with @samcharrington for the @twimlai podcast.
Check it out!
twimlai.com/go/781.
Airbnb is proud to be a Platinum Sponsor of EMNLP 2026!
We'll be sharing talks & posters on LLM tool-agent harnesses, agentic orchestration, and agentic RAG, plus we're hiring applied NLP/AI talent.
Come say hi at our booth and let's explore whats next for AI at Airbnb!
1
2,153
How do you train search before the first real query? 🏡
Airbnb’s CASTLE combines AI with booking behavior to generate realistic search queries and relevance labels, helping natural language search better understand what guests want.
See how it works 👇
medium.com/airbnb-engineerin…
Synthetic benchmarks test what you imagine. Real traffic tests what your database actually handles.
At Airbnb, we replay production traffic offline to catch issues before they hit production.
Check out our latest blog post to learn more!
medium.com/airbnb-engineerin…
How do you design AI products that don't break when LLMs get unpredictable?
Our very own Alicia Bendz will share real lessons, architecture patterns, and the questions every team should ask before shipping at next Wednesday's Women Who Code Summit.
womenwhocode.com/agenda/?bz_…
What if you could personalize an experience without knowing who the user is?
Airbnb's new Proximity Features use collective signals from nearby users to power smart, privacy-safe recommendations with no persistent user identity required.
medium.com/airbnb-engineerin…
7
3,011
Intelligence without methodology is not science 🚨
Here's how we built an agent harness that encodes scientific rigor into every AI-driven investigation: reproducible, auditable, and scalable across any unstructured text.
medium.com/airbnb-engineerin…
3
3,103
Stale assumptions. Ghosts from COVID. Models still forecasting a world that no longer exists. A good forecasting team doesn't just learn from shocks, it learns to forget them, on purpose. Here's how we built the discipline to know when. medium.com/airbnb-engineerin…
What happens when a login screen doesn't know which challenge a user can actually complete? We rebuilt authentication from the ground up and the results speak for themselves. medium.com/airbnb-engineerin…
3
3,472
False confidence. Undetected regressions. Wasted effort. These are what happen when GenAI teams skip rigorous evaluation. We've been there. Here's how we team built eval-driven development to ship trustworthy AI products.
medium.com/airbnb-engineerin…
Airbnb Engineering retweeted
It's earnings season, so the bubble discourse is back. Two things are true at once: we don't have enough compute, and we're overbuilding it. Both camps are sort of right. Here's the game that I think makes that possible.
97.8% of guest activity is just listing views. We built a sequence model that learns from that noise plus years of booking history — and made search meaningfully more personalized.
medium.com/airbnb-engineerin…
Training an LLM is the easy part. Trusting your evaluation of it is hard. We rebuilt our eval infra around deterministic evaluation to minimize confounders + "micro adapters" for targeted fixes & went from weeks long iteration loops to same day fixes. medium.com/airbnb-engineerin…
9
3,295
When we expanded from Homes to Experiences & Services, our data architecture had to scale too. Here's how we built a consistent, flexible data modeling framework for a multi-product world 👇 medium.com/airbnb-engineerin…
4
2,051





