Golden files stop working when the output is different every time. How we build, version and review fixtures for systems whose correct answer is a range rather than a value.
","track":"AI Engineering","format":"Talk (30 min)","room":null,"startsAt":null,"endsAt":null,"speakers":[{"id":"demo-p-2","name":"Dana Kowalski","title":"Principal Engineer","company":"Substrate AI","role":"speaker"}]},{"id":"demo-s-41","code":"SESS-42","title":"From Notebook to Service in One Afternoon","descriptionHtml":"A hands-on session taking a research notebook to a deployed service with tests, monitoring and a rollback path. Aimed at data scientists who have been asked to ship, and at the engineers helping them.
","track":"Developer Experience","format":"Workshop (120 min)","room":null,"startsAt":null,"endsAt":null,"speakers":[{"id":"demo-p-15","name":"Diego Ramirez","title":"Data Platform Engineer","company":"Verdant","role":"speaker"}]},{"id":"ae62f91f-f263-4348-a833-bf6065aa6977","code":"SESS-51","title":"UPDATED: Taming 40-Minute CI: Incremental Builds at Monorepo Scale","descriptionHtml":"Our monorepo CI took 40 minutes on a good day. This talk walks through how we cut it to 6 minutes with content-addressed caching, remote execution, and a test-selection model — including the two migrations that failed first. You'll leave with a decision framework for which incremental-build investments pay off at which repo sizes, and the graphs to convince your platform team. Updated: now includes 2026 benchmark data. This session now includes a live demo of remote build caching.
","track":"Platform & Infra","format":"Talk (30 min)","room":"Room 2A","startsAt":"2026-09-15T16:00:00.000Z","endsAt":"2026-09-15T16:30:00.000Z","speakers":[{"id":"a0171bb2-3c93-44c0-9d44-f8265a64bba4","name":"Priya Raman","title":null,"company":null,"role":"speaker"}]},{"id":"demo-s-0","code":"SESS-1","title":"Context Engineering in Production: What Actually Survives Contact With Users","descriptionHtml":"A year of shipping agent products taught us that context design outlives model choice. This talk covers the patterns that survived: compaction budgets, retrieval boundaries, and the observability you need before you can tune anything.
","track":"AI Engineering","format":"Keynote (45 min)","room":"Room 2A","startsAt":"2026-09-15T16:45:00.000Z","endsAt":"2026-09-15T17:30:00.000Z","speakers":[{"id":"demo-p-0","name":"Priya Raman","title":"Staff Engineer","company":"Latticework Systems","role":"speaker"}]},{"id":"demo-s-2","code":"SESS-3","title":"Your AI Pair Programmer Is Lying to You: Verification Patterns That Scale","descriptionHtml":"Generated code compiles, passes review, and still breaks production. We catalogued a year of agent-written incidents and built verification layers that catch what human review misses — without slowing anyone down.
","track":"AI Engineering","format":"Talk (30 min)","room":"Main Stage","startsAt":"2026-09-15T17:30:00.000Z","endsAt":"2026-09-15T18:00:00.000Z","speakers":[{"id":"demo-p-2","name":"Dana Kowalski","title":"Principal Engineer","company":"Substrate AI","role":"speaker"}]},{"id":"demo-s-4","code":"SESS-5","title":"The Edge Database You Already Have: SQLite Patterns for Global Apps","descriptionHtml":"Read replicas at the edge, single-writer semantics, and what happens when you stop pretending you need a cluster. Real latency numbers from three production migrations, including the one we rolled back.
","track":"Platform & Infra","format":"Talk (30 min)","room":"Room 2A","startsAt":"2026-09-15T17:30:00.000Z","endsAt":"2026-09-15T18:00:00.000Z","speakers":[{"id":"demo-p-4","name":"Elena Vasquez","title":"ML Engineer","company":"Braid Research","role":"speaker"}]},{"id":"demo-s-7","code":"SESS-8","title":"Streaming Structured Output: UX Patterns Beyond the Token Firehose","descriptionHtml":"Users do not want to watch JSON assemble itself. Patterns for progressive disclosure, optimistic rendering, and graceful repair when the model changes its mind halfway through a response.
","track":"Developer Experience","format":"Talk (30 min)","room":"Room 2B","startsAt":"2026-09-15T17:30:00.000Z","endsAt":"2026-09-15T18:00:00.000Z","speakers":[{"id":"demo-p-7","name":"Raj Patel","title":"Founding Engineer","company":"Tandem Agents","role":"speaker"}]},{"id":"demo-s-1","code":"SESS-2","title":"Taming 40-Minute CI: Incremental Builds at Monorepo Scale","descriptionHtml":"Our monorepo CI hit forty minutes and engineers stopped trusting it. We rebuilt it around content-addressed caching and got to six. A decision framework for which incremental-build investments actually pay off, and which ones are cargo cult.
","track":"Platform & Infra","format":"Talk (30 min)","room":"Room 2A","startsAt":"2026-09-15T18:15:00.000Z","endsAt":"2026-09-15T18:45:00.000Z","speakers":[{"id":"demo-p-1","name":"Marcus Okafor","title":"Staff Developer Advocate","company":"Cloudreach Labs","role":"speaker"}]},{"id":"demo-s-3","code":"SESS-4","title":"Multi-Agent Systems Are Just Distributed Systems (And That's Good News)","descriptionHtml":"Everything your team already knows about queues, retries, idempotency and backpressure applies directly to agent orchestration. A tour of production multi-agent failures and the boring distributed-systems fixes that solved each one.
","track":"AI Engineering","format":"Talk (30 min)","room":"Main Stage","startsAt":"2026-09-15T18:15:00.000Z","endsAt":"2026-09-15T18:45:00.000Z","speakers":[{"id":"demo-p-3","name":"Yuki Tanaka","title":"Head of Platform","company":"Kitsune Cloud","role":"speaker"}]},{"id":"demo-s-10","code":"SESS-11","title":"Design Systems for AI Products: Calm Interfaces for Chaotic Models","descriptionHtml":"Non-determinism does not excuse jittery interfaces. Tokens, motion budgets and content design for products where the backend is a probability distribution and the user still needs to feel in control.
","track":"Developer Experience","format":"Talk (30 min)","room":"Room 2B","startsAt":"2026-09-15T18:15:00.000Z","endsAt":"2026-09-15T18:45:00.000Z","speakers":[{"id":"demo-p-10","name":"Grace Liu","title":"AI Product Lead","company":"Meridian","role":"speaker"}]},{"id":"demo-s-6","code":"SESS-7","title":"Developer Platforms People Actually Adopt: Lessons From Three Failures","descriptionHtml":"We built three internal platforms nobody used before one that stuck. The difference was never features — it was migration cost, escape hatches, and letting teams keep their own deploy buttons.
","track":"Developer Experience","format":"Talk (30 min)","room":"Room 2B","startsAt":"2026-09-15T20:15:00.000Z","endsAt":"2026-09-15T20:45:00.000Z","speakers":[{"id":"demo-p-6","name":"Sofia Lindqvist","title":"Engineering Manager","company":"Norrsken Data","role":"speaker"}]},{"id":"demo-s-8","code":"SESS-9","title":"Prompt Injection Is a Solved Problem (If You Do These Five Things)","descriptionHtml":"Provocative title, sober content: capability scoping, output mediation, and the three-layer defence we run in production. Includes live demos of attacks that still work against naive agents in 2026.
","track":"AI Engineering","format":"Talk (30 min)","room":"Main Stage","startsAt":"2026-09-15T20:15:00.000Z","endsAt":"2026-09-15T20:45:00.000Z","speakers":[{"id":"demo-p-8","name":"Amara Diallo","title":"Security Researcher","company":"Halcyon Labs","role":"speaker"}]},{"id":"demo-s-23","code":"SESS-24","title":"Ten Wrong Turns Migrating a Data Platform","descriptionHtml":"A fast tour of the ten decisions we would make differently, from partition keys to the migration window we scheduled during a holiday. Ten minutes, ten mistakes, no filler.
","track":"Platform & Infra","format":"Lightning Talk (10 min)","room":"Room 2A","startsAt":"2026-09-15T20:15:00.000Z","endsAt":"2026-09-15T20:25:00.000Z","speakers":[{"id":"demo-p-15","name":"Diego Ramirez","title":"Data Platform Engineer","company":"Verdant","role":"speaker"}]},{"id":"demo-s-24","code":"SESS-25","title":"Reading Eval Traces Like a Detective","descriptionHtml":"A short, practical walkthrough of how to read a failing evaluation trace: what to look at first, which signals lie, and how to turn one bad output into a reproducible test case.
","track":"AI Engineering","format":"Lightning Talk (10 min)","room":"Room 2A","startsAt":"2026-09-15T20:35:00.000Z","endsAt":"2026-09-15T20:45:00.000Z","speakers":[{"id":"demo-p-4","name":"Elena Vasquez","title":"ML Engineer","company":"Braid Research","role":"speaker"}]},{"id":"demo-s-36","code":"SESS-37","title":"Panel: What Breaks When AI Products Meet Real Users","descriptionHtml":"Four practitioners on the gap between the demo and the deployment: the failures that only appear at scale, the support burden nobody budgets for, and what they would build differently. Audience questions from the start.
","track":"AI Engineering","format":"Panel (45 min)","room":"Main Stage","startsAt":"2026-09-15T22:15:00.000Z","endsAt":"2026-09-15T23:00:00.000Z","speakers":[{"id":"demo-p-10","name":"Grace Liu","title":"AI Product Lead","company":"Meridian","role":"moderator"},{"id":"demo-p-5","name":"Tom Achebe","title":"CTO","company":"Forgeline","role":"panelist"},{"id":"demo-p-13","name":"Kwame Mensah","title":"Director of Engineering","company":"Northwind Health","role":"panelist"},{"id":"demo-p-20","name":"Aisha Rahman","title":"VP Engineering","company":"Loomis","role":"panelist"}]},{"id":"demo-s-21","code":"SESS-22","title":"Human in the Loop, For Real: Clinical AI Under Review","descriptionHtml":"Every model output in our product is read by a clinician before it reaches a patient chart. This keynote covers what that constraint taught us about interface design, latency budgets, and the difference between assistance and automation.
","track":"AI Engineering","format":"Keynote (45 min)","room":"Main Stage","startsAt":"2026-09-16T16:30:00.000Z","endsAt":"2026-09-16T17:15:00.000Z","speakers":[{"id":"demo-p-13","name":"Kwame Mensah","title":"Director of Engineering","company":"Northwind Health","role":"speaker"}]},{"id":"demo-s-5","code":"SESS-6","title":"Evals Are Your Only Moat: Building Test Suites for LLM Products","descriptionHtml":"Model upgrades break prompts silently. In this hands-on workshop you will build an eval pipeline from scratch: rubric design, failure taxonomies, CI wiring, and the regression report that doubles as your product spec. Bring a laptop.
","track":"AI Engineering","format":"Workshop (120 min)","room":"Workshop Lab","startsAt":"2026-09-16T17:30:00.000Z","endsAt":"2026-09-16T19:30:00.000Z","speakers":[{"id":"demo-p-5","name":"Tom Achebe","title":"CTO","company":"Forgeline","role":"speaker"},{"id":"demo-p-16","name":"Hana Sato","title":"Research Engineer","company":"Aperture AI","role":"speaker"}]},{"id":"demo-s-11","code":"SESS-12","title":"Postgres to the Edge: A Migration Post-Mortem","descriptionHtml":"What broke, what got faster, and the three queries that nearly made us roll the whole thing back. An honest accounting with p99 charts and the cost line nobody put in the original proposal.
","track":"Platform & Infra","format":"Talk (30 min)","room":"Room 2A","startsAt":"2026-09-16T17:30:00.000Z","endsAt":"2026-09-16T18:00:00.000Z","speakers":[{"id":"demo-p-11","name":"Omar Haddad","title":"Distinguished Engineer","company":"Qanat Systems","role":"speaker"}]},{"id":"demo-s-20","code":"SESS-21","title":"On Call for a Model: Reliability Practices for Inference Platforms","descriptionHtml":"Inference platforms fail in ways your existing runbooks do not describe. We rebuilt our alerting around output quality rather than uptime, and the pager got quieter and more useful at the same time.
","track":"Platform & Infra","format":"Talk (30 min)","room":"Main Stage","startsAt":"2026-09-16T17:30:00.000Z","endsAt":"2026-09-16T18:00:00.000Z","speakers":[{"id":"demo-p-12","name":"Nina Berg","title":"Staff SRE","company":"Halberd","role":"speaker"}]},{"id":"demo-s-22","code":"SESS-23","title":"Undo Is a Feature: Designing Reversibility Into AI Products","descriptionHtml":"Confidence in an AI product comes from being able to take it back. A design tour of undo, diffing and preview patterns, and the data model changes each one demands.
","track":"Developer Experience","format":"Talk (30 min)","room":"Room 2B","startsAt":"2026-09-16T18:15:00.000Z","endsAt":"2026-09-16T18:45:00.000Z","speakers":[{"id":"demo-p-14","name":"Ines Moreau","title":"Design Lead","company":"Calyx","role":"speaker"}]},{"id":"demo-s-25","code":"SESS-26","title":"Your Legacy Systems Are the Hard Part","descriptionHtml":"Most AI integration projects stall on a mainframe export, not a model. Patterns for wrapping systems you cannot change, and how to negotiate data contracts with teams that have no incentive to help you.
","track":"Platform & Infra","format":"Talk (30 min)","room":"Main Stage","startsAt":"2026-09-16T18:15:00.000Z","endsAt":"2026-09-16T18:45:00.000Z","speakers":[{"id":"demo-p-17","name":"Ben Oyelaran","title":"Principal Architect","company":"Steelhead","role":"speaker"}]},{"id":"demo-s-27","code":"SESS-28","title":"Scheduling GPU Work Without a Scheduler Team","descriptionHtml":"Fair queuing, preemption and priority for a fleet you cannot afford to leave idle — implemented by two engineers with a job queue and a lot of restraint about what to build.
","track":"Platform & Infra","format":"Talk (30 min)","room":"Room 2A","startsAt":"2026-09-16T18:15:00.000Z","endsAt":"2026-09-16T18:45:00.000Z","speakers":[{"id":"demo-p-19","name":"Theo Whitfield","title":"Staff Engineer","company":"Junction","role":"speaker"}]},{"id":"demo-s-28","code":"SESS-29","title":"Growing an Engineering Org Around an AI Product","descriptionHtml":"We went from twelve engineers to ninety while shipping our first AI product, and got a lot of it wrong. Hiring, team topology, and the two reorganisations we could have skipped.
","track":"Developer Experience","format":"Talk (30 min)","room":"Main Stage","startsAt":"2026-09-16T20:15:00.000Z","endsAt":"2026-09-16T20:45:00.000Z","speakers":[{"id":"demo-p-20","name":"Aisha Rahman","title":"VP Engineering","company":"Loomis","role":"speaker"}]},{"id":"demo-s-32","code":"SESS-33","title":"Deleting Internal Tools: A Year of Saying No","descriptionHtml":"We removed nineteen internal tools in twelve months and replaced them with four. How we chose, how we migrated people who did not want to move, and the two deletions we regret.
","track":"Developer Experience","format":"Talk (30 min)","room":"Room 2B","startsAt":"2026-09-16T20:15:00.000Z","endsAt":"2026-09-16T20:45:00.000Z","speakers":[{"id":"demo-p-24","name":"Zoe Kimura","title":"Engineering Manager","company":"Fernbrook","role":"speaker"}]},{"id":"demo-s-46","code":"SESS-47","title":"Cutting Our Inference Bill in Half Without Touching Quality","descriptionHtml":"Batching, caching and routing decisions that halved our spend, plus the two optimisations that looked great on a dashboard and cost us accuracy. Numbers from a fleet running around the clock.
","track":"Platform & Infra","format":"Talk (30 min)","room":"Main Stage","startsAt":"2026-09-16T21:15:00.000Z","endsAt":"2026-09-16T21:45:00.000Z","speakers":[{"id":"demo-p-9","name":"Felix Braun","title":"Infrastructure Lead","company":"Morgenrot","role":"speaker"}]},{"id":"demo-s-47","code":"SESS-48","title":"The Support Queue Is Your Best Eval Set","descriptionHtml":"Every complaint is a labelled failure somebody already wrote down for you. How we turn support tickets into evaluation cases weekly, and what that changed about which bugs we fix first.
","track":"AI Engineering","format":"Talk (30 min)","room":"Room 2A","startsAt":"2026-09-16T21:15:00.000Z","endsAt":"2026-09-16T21:45:00.000Z","speakers":[{"id":"demo-p-22","name":"Leila Farsi","title":"Staff Product Manager","company":"Coastline","role":"speaker"}]},{"id":"demo-s-48","code":"SESS-49","title":"Closing Keynote: The Boring Future of AI Engineering","descriptionHtml":"A bet on where this goes: fewer demos, more contracts; fewer frameworks, more schemas; and the unglamorous integration work that will decide which products are still running in five years.
","track":"AI Engineering","format":"Keynote (45 min)","room":"Main Stage","startsAt":"2026-09-16T22:15:00.000Z","endsAt":"2026-09-16T23:00:00.000Z","speakers":[{"id":"demo-p-17","name":"Ben Oyelaran","title":"Principal Architect","company":"Steelhead","role":"speaker"}]},{"id":"demo-s-13","code":"SESS-14","title":"The Accessibility Debt of AI Interfaces","descriptionHtml":"Screen readers meet streaming text. A practical audit of the AI interface patterns that have quietly become standard, which assistive technologies they break, and the fixes that make them usable for everyone.
","track":"Developer Experience","format":"Talk (30 min)","room":"Main Stage","startsAt":"2026-09-17T16:45:00.000Z","endsAt":"2026-09-17T17:15:00.000Z","speakers":[{"id":"demo-p-14","name":"Ines Moreau","title":"Design Lead","company":"Calyx","role":"speaker"}]},{"id":"demo-s-12","code":"SESS-13","title":"Fine-Tuning Is Back: Small Models for Narrow Jobs","descriptionHtml":"GPU prices fell and our inference bill did not. Where task-specific small models beat frontier APIs, how to run that comparison honestly, and the four jobs where fine-tuning is still the wrong answer.
","track":"AI Engineering","format":"Talk (30 min)","room":"Workshop Lab","startsAt":"2026-09-17T17:15:00.000Z","endsAt":"2026-09-17T17:45:00.000Z","speakers":[{"id":"demo-p-16","name":"Hana Sato","title":"Research Engineer","company":"Aperture AI","role":"speaker"}]},{"id":"demo-s-26","code":"SESS-27","title":"Workshops That Don't Waste an Afternoon","descriptionHtml":"A working session for anyone who runs technical workshops: pacing, environment setup that survives conference wifi, exercise design, and how to rescue a room that has fallen behind. You will leave with a rewritten outline for your own workshop.
","track":"Developer Experience","format":"Workshop (120 min)","room":"Room 2A","startsAt":"2026-09-17T17:15:00.000Z","endsAt":"2026-09-17T19:15:00.000Z","speakers":[{"id":"demo-p-18","name":"Clara Novak","title":"Head of Developer Relations","company":"Ambit","role":"speaker"}]},{"id":"demo-s-30","code":"SESS-31","title":"Writing the Safety Spec Before the Feature","descriptionHtml":"How our product team writes the abuse cases, refusal behaviour and escalation path before a line of code exists, and what it costs when we skip it. Includes the template we use.
","track":"AI Engineering","format":"Talk (30 min)","room":"Main Stage","startsAt":"2026-09-17T17:30:00.000Z","endsAt":"2026-09-17T18:00:00.000Z","speakers":[{"id":"demo-p-22","name":"Leila Farsi","title":"Staff Product Manager","company":"Coastline","role":"speaker"}]},{"id":"demo-s-35","code":"SESS-36","title":"The Cost of a Token: FinOps for LLM Products","descriptionHtml":"Unit economics for products where the marginal cost is real. Attribution, caching strategy, model routing, and the dashboard that let us cut spend by sixty percent without a quality regression.
","track":"Platform & Infra","format":"Talk (30 min)","room":"Main Stage","startsAt":"2026-09-17T18:15:00.000Z","endsAt":"2026-09-17T18:45:00.000Z","speakers":[{"id":"demo-p-11","name":"Omar Haddad","title":"Distinguished Engineer","company":"Qanat Systems","role":"speaker"}]},{"id":"demo-s-31","code":"SESS-32","title":"Keeping 3,000 Demos Online: Event Networks Under Load","descriptionHtml":"Ten minutes on what a conference network actually has to survive, why the failure is almost never bandwidth, and the three changes that made ours stop falling over.
","track":"Platform & Infra","format":"Lightning Talk (10 min)","room":"Workshop Lab","startsAt":"2026-09-17T18:30:00.000Z","endsAt":"2026-09-17T18:40:00.000Z","speakers":[{"id":"demo-p-23","name":"Marco Bianchi","title":"Site Reliability Engineer","company":"Palazzo","role":"speaker"}]},{"id":"demo-s-29","code":"SESS-30","title":"Compiling Models to the Edge: A Toolchain Tour","descriptionHtml":"Quantisation, kernel selection and the compiler passes that decide whether your model fits the device. A tour of where the time actually goes, with benchmarks you can reproduce.
","track":"Platform & Infra","format":"Talk (30 min)","room":"Room 2A","startsAt":"2026-09-17T19:30:00.000Z","endsAt":"2026-09-17T20:00:00.000Z","speakers":[{"id":"demo-p-21","name":"Ivan Petrov","title":"Compiler Engineer","company":"Northgate","role":"speaker"}]},{"id":"demo-s-33","code":"SESS-34","title":"Buying AI in a Big Company: A Field Guide","descriptionHtml":"Procurement, security review, and the eighteen-month contract cycle — from the perspective of the engineer who has to make something work inside it. Practical advice for both buyers and vendors.
","track":"Developer Experience","format":"Talk (30 min)","room":"Room 2B","startsAt":"2026-09-17T21:00:00.000Z","endsAt":"2026-09-17T21:30:00.000Z","speakers":[{"id":"demo-p-25","name":"Andre Lucas","title":"Head of AI","company":"Bellweather","role":"speaker"}]}],"speakers":[{"id":"demo-p-5","name":"Tom Achebe","title":"CTO","company":"Forgeline","bio":"Tom co-founded Forgeline after leading platform engineering at two payments companies. He is the reason Forgeline had an evaluation suite before it had a marketing site.","photoUrl":"/files/demo-photo-5","sessions":[{"id":"demo-s-36","title":"Panel: What Breaks When AI Products Meet Real Users","startsAt":"2026-09-15T22:15:00.000Z","room":"Main Stage"},{"id":"demo-s-5","title":"Evals Are Your Only Moat: Building Test Suites for LLM Products","startsAt":"2026-09-16T17:30:00.000Z","room":"Workshop Lab"}]},{"id":"demo-p-12","name":"Nina Berg","title":"Staff SRE","company":"Halberd","bio":"Nina is on call for a platform that serves inference in forty countries. She thinks most reliability work is really vocabulary work: naming failure modes precisely enough to fix them.","photoUrl":"/files/demo-photo-12","sessions":[{"id":"demo-s-20","title":"On Call for a Model: Reliability Practices for Inference Platforms","startsAt":"2026-09-16T17:30:00.000Z","room":"Main Stage"}]},{"id":"demo-p-23","name":"Marco Bianchi","title":"Site Reliability Engineer","company":"Palazzo","bio":"Marco keeps Palazzo's event infrastructure standing through traffic spikes measured in orders of magnitude. He has run the network for three conferences and lived to describe it.","photoUrl":"/files/demo-photo-23","sessions":[{"id":"demo-s-31","title":"Keeping 3,000 Demos Online: Event Networks Under Load","startsAt":"2026-09-17T18:30:00.000Z","room":"Workshop Lab"}]},{"id":"demo-p-9","name":"Felix Braun","title":"Infrastructure Lead","company":"Morgenrot","bio":"Felix keeps Morgenrot's GPU fleet busy and its finance team calm. He wrote two internal post-mortems that are now required onboarding reading.","photoUrl":"/files/demo-photo-9","sessions":[{"id":"demo-s-46","title":"Cutting Our Inference Bill in Half Without Touching Quality","startsAt":"2026-09-16T21:15:00.000Z","room":"Main Stage"}]},{"id":"demo-p-8","name":"Amara Diallo","title":"Security Researcher","company":"Halcyon Labs","bio":"Amara leads red-team work on agent systems at Halcyon. Her disclosure of the tool-confusion class of prompt injections is why several vendors now scope tool permissions per call.","photoUrl":"/files/demo-photo-8","sessions":[{"id":"demo-s-8","title":"Prompt Injection Is a Solved Problem (If You Do These Five Things)","startsAt":"2026-09-15T20:15:00.000Z","room":"Main Stage"}]},{"id":"demo-p-22","name":"Leila Farsi","title":"Staff Product Manager","company":"Coastline","bio":"Leila owns the trust and safety surface for Coastline's assistant. She used to be a reporter, which shows in the way she writes specifications.","photoUrl":"/files/demo-photo-22","sessions":[{"id":"demo-s-47","title":"The Support Queue Is Your Best Eval Set","startsAt":"2026-09-16T21:15:00.000Z","room":"Room 2A"},{"id":"demo-s-30","title":"Writing the Safety Spec Before the Feature","startsAt":"2026-09-17T17:30:00.000Z","room":"Main Stage"}]},{"id":"demo-p-11","name":"Omar Haddad","title":"Distinguished Engineer","company":"Qanat Systems","bio":"Omar has spent twenty years on data infrastructure, most recently the query layer behind Qanat's edge platform. He is unreasonably interested in what happens at p99.9.","photoUrl":"/files/demo-photo-11","sessions":[{"id":"demo-s-11","title":"Postgres to the Edge: A Migration Post-Mortem","startsAt":"2026-09-16T17:30:00.000Z","room":"Room 2A"},{"id":"demo-s-35","title":"The Cost of a Token: FinOps for LLM Products","startsAt":"2026-09-17T18:15:00.000Z","room":"Main Stage"}]},{"id":"demo-p-24","name":"Zoe Kimura","title":"Engineering Manager","company":"Fernbrook","bio":"Zoe leads the platform team at Fernbrook and spent the past year replacing internal tools with fewer, better ones. She keeps a public list of everything she deleted.","photoUrl":"/files/demo-photo-24","sessions":[{"id":"demo-s-32","title":"Deleting Internal Tools: A Year of Saying No","startsAt":"2026-09-16T20:15:00.000Z","room":"Room 2B"}]},{"id":"demo-p-2","name":"Dana Kowalski","title":"Principal Engineer","company":"Substrate AI","bio":"Dana runs the code-generation platform at Substrate, where about a third of merged pull requests are now machine-authored. Her work is mostly the verification layer that makes that survivable.","photoUrl":"/files/demo-photo-2","sessions":[{"id":"demo-s-37","title":"Test Data Is a Product: Fixtures for Nondeterministic Systems","startsAt":null,"room":null},{"id":"demo-s-2","title":"Your AI Pair Programmer Is Lying to You: Verification Patterns That Scale","startsAt":"2026-09-15T17:30:00.000Z","room":"Main Stage"}]},{"id":"demo-p-6","name":"Sofia Lindqvist","title":"Engineering Manager","company":"Norrsken Data","bio":"Sofia manages the developer-experience group at Norrsken, a team she started by asking every engineer which internal tool they would delete first. She has been an SRE, a compiler engineer and, briefly, a professional oboist.","photoUrl":"/files/demo-photo-6","sessions":[{"id":"demo-s-6","title":"Developer Platforms People Actually Adopt: Lessons From Three Failures","startsAt":"2026-09-15T20:15:00.000Z","room":"Room 2B"}]},{"id":"demo-p-10","name":"Grace Liu","title":"AI Product Lead","company":"Meridian","bio":"Grace has shipped AI features to twelve million users and killed about as many that never earned their place. She writes about interface design for systems that answer differently every time.","photoUrl":"/files/demo-photo-10","sessions":[{"id":"demo-s-10","title":"Design Systems for AI Products: Calm Interfaces for Chaotic Models","startsAt":"2026-09-15T18:15:00.000Z","room":"Room 2B"},{"id":"demo-s-36","title":"Panel: What Breaks When AI Products Meet Real Users","startsAt":"2026-09-15T22:15:00.000Z","room":"Main Stage"}]},{"id":"demo-p-25","name":"Andre Lucas","title":"Head of AI","company":"Bellweather","bio":"Andre runs a ten-person AI group inside a four-thousand-person company and spends half his time on procurement. He is honest about which of those two is harder.","photoUrl":"/files/demo-photo-25","sessions":[{"id":"demo-s-33","title":"Buying AI in a Big Company: A Field Guide","startsAt":"2026-09-17T21:00:00.000Z","room":"Room 2B"}]},{"id":"demo-p-13","name":"Kwame Mensah","title":"Director of Engineering","company":"Northwind Health","bio":"Kwame leads the clinical AI group at Northwind, where every model output is reviewed by a clinician before it reaches a chart. He is candid about what regulation does and does not actually slow down.","photoUrl":"/files/demo-photo-13","sessions":[{"id":"demo-s-36","title":"Panel: What Breaks When AI Products Meet Real Users","startsAt":"2026-09-15T22:15:00.000Z","room":"Main Stage"},{"id":"demo-s-21","title":"Human in the Loop, For Real: Clinical AI Under Review","startsAt":"2026-09-16T16:30:00.000Z","room":"Main Stage"}]},{"id":"demo-p-14","name":"Ines Moreau","title":"Design Lead","company":"Calyx","bio":"Ines designed Calyx's assistant interface and then its undo system, in that order of difficulty. She is a recovering print designer who still sets type by hand for fun.","photoUrl":"/files/demo-photo-14","sessions":[{"id":"demo-s-22","title":"Undo Is a Feature: Designing Reversibility Into AI Products","startsAt":"2026-09-16T18:15:00.000Z","room":"Room 2B"},{"id":"demo-s-13","title":"The Accessibility Debt of AI Interfaces","startsAt":"2026-09-17T16:45:00.000Z","room":"Main Stage"}]},{"id":"demo-p-18","name":"Clara Novak","title":"Head of Developer Relations","company":"Ambit","bio":"Clara grew Ambit's developer community from a chat server with nine people in it. She has run more than forty workshops and has firm opinions about the ones that do not work.","photoUrl":"/files/demo-photo-18","sessions":[{"id":"demo-s-26","title":"Workshops That Don't Waste an Afternoon","startsAt":"2026-09-17T17:15:00.000Z","room":"Room 2A"}]},{"id":"demo-p-1","name":"Marcus Okafor","title":"Staff Developer Advocate","company":"Cloudreach Labs","bio":"Marcus spent six years on build infrastructure before moving into advocacy, and still maintains the caching tool that came out of it. Most of his talks open with a CI pipeline somebody let rot for two years.","photoUrl":"/files/demo-photo-1","sessions":[{"id":"demo-s-1","title":"Taming 40-Minute CI: Incremental Builds at Monorepo Scale","startsAt":"2026-09-15T18:15:00.000Z","room":"Room 2A"}]},{"id":"demo-p-17","name":"Ben Oyelaran","title":"Principal Architect","company":"Steelhead","bio":"Ben designs integration platforms for companies with more legacy systems than engineers. He believes most AI adoption problems are really data-contract problems wearing a new hat.","photoUrl":"/files/demo-photo-17","sessions":[{"id":"demo-s-25","title":"Your Legacy Systems Are the Hard Part","startsAt":"2026-09-16T18:15:00.000Z","room":"Main Stage"},{"id":"demo-s-48","title":"Closing Keynote: The Boring Future of AI Engineering","startsAt":"2026-09-16T22:15:00.000Z","room":"Main Stage"}]},{"id":"demo-p-7","name":"Raj Patel","title":"Founding Engineer","company":"Tandem Agents","bio":"Raj was employee number two at Tandem and wrote the streaming layer the product is now known for. He has rewritten it three times and expects a fourth.","photoUrl":"/files/demo-photo-7","sessions":[{"id":"demo-s-7","title":"Streaming Structured Output: UX Patterns Beyond the Token Firehose","startsAt":"2026-09-15T17:30:00.000Z","room":"Room 2B"}]},{"id":"demo-p-21","name":"Ivan Petrov","title":"Compiler Engineer","company":"Northgate","bio":"Ivan works on the toolchain that compiles Northgate's models down to edge targets. He is happiest when a benchmark gets slower and he gets to find out why.","photoUrl":"/files/demo-photo-21","sessions":[{"id":"demo-s-29","title":"Compiling Models to the Edge: A Toolchain Tour","startsAt":"2026-09-17T19:30:00.000Z","room":"Room 2A"}]},{"id":"demo-p-20","name":"Aisha Rahman","title":"VP Engineering","company":"Loomis","bio":"Aisha grew Loomis engineering from twelve people to ninety while shipping the company's first AI product. She writes mostly about the parts of that which went badly.","photoUrl":"/files/demo-photo-20","sessions":[{"id":"demo-s-36","title":"Panel: What Breaks When AI Products Meet Real Users","startsAt":"2026-09-15T22:15:00.000Z","room":"Main Stage"},{"id":"demo-s-28","title":"Growing an Engineering Org Around an AI Product","startsAt":"2026-09-16T20:15:00.000Z","room":"Main Stage"}]},{"id":"a0171bb2-3c93-44c0-9d44-f8265a64bba4","name":"Priya Raman","title":null,"company":null,"bio":"Priya Raman is a Principal Engineer at Latticework Systems where she leads the build-tooling platform team. She previously maintained the open-source task runner 'gantry' and has spoken at over a dozen developer conferences on build systems, CI reliability, and developer productivity metrics.","photoUrl":null,"sessions":[{"id":"ae62f91f-f263-4348-a833-bf6065aa6977","title":"UPDATED: Taming 40-Minute CI: Incremental Builds at Monorepo Scale","startsAt":"2026-09-15T16:00:00.000Z","room":"Room 2A"}]},{"id":"demo-p-0","name":"Priya Raman","title":"Staff Engineer","company":"Latticework Systems","bio":"Priya leads the agent platform team at Latticework, where she rebuilt the company's retrieval stack around explicit context budgets after a rewrite that took two attempts. She writes the Context Notes newsletter and has been shipping LLM products since the first public API keys.","photoUrl":"/files/demo-photo-0","sessions":[{"id":"demo-s-0","title":"Context Engineering in Production: What Actually Survives Contact With Users","startsAt":"2026-09-15T16:45:00.000Z","room":"Room 2A"}]},{"id":"demo-p-15","name":"Diego Ramirez","title":"Data Platform Engineer","company":"Verdant","bio":"Diego moved Verdant's analytics stack off a warehouse it had outgrown and wrote down every wrong turn on the way. He mentors at two Bay Area programs.","photoUrl":"/files/demo-photo-15","sessions":[{"id":"demo-s-41","title":"From Notebook to Service in One Afternoon","startsAt":null,"room":null},{"id":"demo-s-23","title":"Ten Wrong Turns Migrating a Data Platform","startsAt":"2026-09-15T20:15:00.000Z","room":"Room 2A"}]},{"id":"demo-p-16","name":"Hana Sato","title":"Research Engineer","company":"Aperture AI","bio":"Hana works on small-model distillation and spends most of her week reading evaluation traces. She was a computational linguist before that and it shows in her error taxonomies.","photoUrl":"/files/demo-photo-16","sessions":[{"id":"demo-s-5","title":"Evals Are Your Only Moat: Building Test Suites for LLM Products","startsAt":"2026-09-16T17:30:00.000Z","room":"Workshop Lab"},{"id":"demo-s-12","title":"Fine-Tuning Is Back: Small Models for Narrow Jobs","startsAt":"2026-09-17T17:15:00.000Z","room":"Workshop Lab"}]},{"id":"demo-p-3","name":"Yuki Tanaka","title":"Head of Platform","company":"Kitsune Cloud","bio":"Yuki built Kitsune's multi-region control plane and now leads the team that keeps it boring. Before that: eight years of distributed databases and one very educational New Year's Eve outage.","photoUrl":"/files/demo-photo-3","sessions":[{"id":"demo-s-3","title":"Multi-Agent Systems Are Just Distributed Systems (And That's Good News)","startsAt":"2026-09-15T18:15:00.000Z","room":"Main Stage"}]},{"id":"demo-p-4","name":"Elena Vasquez","title":"ML Engineer","company":"Braid Research","bio":"Elena works on retrieval and evaluation at Braid. She published an open benchmark suite after getting tired of numbers that never reproduced on her own hardware.","photoUrl":"/files/demo-photo-4","sessions":[{"id":"demo-s-4","title":"The Edge Database You Already Have: SQLite Patterns for Global Apps","startsAt":"2026-09-15T17:30:00.000Z","room":"Room 2A"},{"id":"demo-s-24","title":"Reading Eval Traces Like a Detective","startsAt":"2026-09-15T20:35:00.000Z","room":"Room 2A"}]},{"id":"demo-p-19","name":"Theo Whitfield","title":"Staff Engineer","company":"Junction","bio":"Theo works on Junction's orchestration engine and the scheduler underneath it. He came to distributed systems from computational biology and has never entirely left.","photoUrl":"/files/demo-photo-19","sessions":[{"id":"demo-s-27","title":"Scheduling GPU Work Without a Scheduler Team","startsAt":"2026-09-16T18:15:00.000Z","room":"Room 2A"}]}]}