BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//LatentShift//latentshift.ai schedule//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
X-WR-CALNAME:LatentShift ed.01 · program
X-WR-TIMEZONE:Europe/Istanbul
REFRESH-INTERVAL;VALUE=DURATION:PT12H
X-PUBLISHED-TTL:PT12H
BEGIN:VEVENT
UID:ed01-registration-0845@latentshift.ai
DTSTAMP:20261007T125617Z
DTSTART:20261017T054500Z
DTEND:20261017T064500Z
SUMMARY:Registration & coffee
LOCATION:Bağlarbaşı Kongre ve Kültür Merkezi\, Boğaziçi Salonu\, Sel
 ami Ali Mah.\, Gazi Cad. No:22\, 34664 Üsküdar\, İstanbul
GEO:41.024366;29.033571
URL:https://latentshift.ai/schedule/
CATEGORIES:LatentShift
STATUS:CONFIRMED
TRANSP:OPAQUE
END:VEVENT
BEGIN:VEVENT
UID:ed01-opening-0945@latentshift.ai
DTSTAMP:20261007T125617Z
DTSTART:20261017T064500Z
DTEND:20261017T070000Z
SUMMARY:Opening remarks
DESCRIPTION:LatentShift organizers: Bilge Gündüz\, Ayşe Aktağ\, Mehmet 
 İnce\n\nhttps://latentshift.ai/schedule/
LOCATION:Bağlarbaşı Kongre ve Kültür Merkezi\, Boğaziçi Salonu\, Sel
 ami Ali Mah.\, Gazi Cad. No:22\, 34664 Üsküdar\, İstanbul
GEO:41.024366;29.033571
URL:https://latentshift.ai/schedule/
CATEGORIES:LatentShift
STATUS:CONFIRMED
TRANSP:OPAQUE
END:VEVENT
BEGIN:VEVENT
UID:ed01-your-agent-needs-a-home@latentshift.ai
DTSTAMP:20261007T125617Z
DTSTART:20261017T070000Z
DTEND:20261017T073000Z
SUMMARY:Your Agent Needs A Home · Jigar Mody & Bilge Gündüz
DESCRIPTION:Keynote · 30 min · English\n\nJigar Mody · Chief AI Officer\
 , EnterpriseDB\nBilge Gündüz · Staff Machine Learning Engineer\, Enterp
 riseDB\n\nOnly 5% of enterprise AI projects reach production. The agents t
 hat do work are expensive\, because every reasoning step and tool call cos
 ts tokens. Most teams add one new system for memory\, another for context 
 and a third for safety\, and every new system adds cost\, latency and risk
 . We think agents already have a better place to live\, and most teams ove
 rlook it. In this talk\, we show where that place is and how it makes agen
 ts effective\, cheaper to run and governed by default. We will share what 
 we built\, what we measured\, and what is still an open problem.\n\nhttps:
 //latentshift.ai/schedule/your-agent-needs-a-home/
LOCATION:Bağlarbaşı Kongre ve Kültür Merkezi\, Boğaziçi Salonu\, Sel
 ami Ali Mah.\, Gazi Cad. No:22\, 34664 Üsküdar\, İstanbul
GEO:41.024366;29.033571
URL:https://latentshift.ai/schedule/your-agent-needs-a-home/
CATEGORIES:LatentShift,Keynote
STATUS:CONFIRMED
TRANSP:OPAQUE
END:VEVENT
BEGIN:VEVENT
UID:ed01-self-healing-a-gpu-fleet-at-scale@latentshift.ai
DTSTAMP:20261007T125617Z
DTSTART:20261017T073500Z
DTEND:20261017T075500Z
SUMMARY:Self-Healing a GPU Fleet at Scale · Daglar Berk Erdem
DESCRIPTION:Short talk · 20 min · Turkish\n\nDaglar Berk Erdem · Platfor
 m Engineer\, FAL.ai\n\nTurkish Sign Language (TİD) interpretation.\n\nRun
 ning thousands of GPUs means hardware fails constantly: GPUs fall off the 
 PCIe bus\, memory corrupts\, nodes drop off the network. Our fleet runs on
  custom scheduler rather than Kubernetes\, so there is no managed platform
  to auto-repair nodes - we built our own. In this session you will learn h
 ow we detect failures from metrics\, automatically drain and remediate nod
 es with Ansible\, and the safeguards that keep automation from making an o
 utage worse.\n\nhttps://latentshift.ai/schedule/self-healing-a-gpu-fleet-a
 t-scale/
LOCATION:Bağlarbaşı Kongre ve Kültür Merkezi\, Boğaziçi Salonu\, Sel
 ami Ali Mah.\, Gazi Cad. No:22\, 34664 Üsküdar\, İstanbul
GEO:41.024366;29.033571
URL:https://latentshift.ai/schedule/self-healing-a-gpu-fleet-at-scale/
CATEGORIES:LatentShift,Short talk
STATUS:CONFIRMED
TRANSP:OPAQUE
END:VEVENT
BEGIN:VEVENT
UID:ed01-break-1100@latentshift.ai
DTSTAMP:20261007T125617Z
DTSTART:20261017T080000Z
DTEND:20261017T082000Z
SUMMARY:Break
LOCATION:Bağlarbaşı Kongre ve Kültür Merkezi\, Boğaziçi Salonu\, Sel
 ami Ali Mah.\, Gazi Cad. No:22\, 34664 Üsküdar\, İstanbul
GEO:41.024366;29.033571
URL:https://latentshift.ai/schedule/
CATEGORIES:LatentShift
STATUS:CONFIRMED
TRANSP:OPAQUE
END:VEVENT
BEGIN:VEVENT
UID:ed01-rlaif-ile-model-finetuning@latentshift.ai
DTSTAMP:20261007T125617Z
DTSTART:20261017T082000Z
DTEND:20261017T090000Z
SUMMARY:Removing the Human Bottleneck in Model Fine-Tuning/Pre-Training wit
 h RLAIF · Alican Kiraz
DESCRIPTION:Long talk · 40 min · Turkish\n\nAlican Kiraz · Sr. Staff Eng
 ineer - AI/ML & Agentic AI\, Trendyol\n\nTurkish Sign Language (TİD) inte
 rpretation.\n\nWith Reinforcement Learning from AI Feedback (RLAIF)\, we c
 an now create Instruct- and SFT-based reinforcement learning loops for mod
 els with minimal reliance on human data. This allows models to train witho
 ut human involvement in a teacher/student structure\, without requiring hu
 man feedback. With the setup I built in my own homelab\, I can develop my 
 Turkish-language open-source LLM and VLM models for finance\, law\, and cy
 bersecurity 24/7\, with minimal human interaction.\n\nhttps://latentshift.
 ai/schedule/rlaif-ile-model-finetuning/
LOCATION:Bağlarbaşı Kongre ve Kültür Merkezi\, Boğaziçi Salonu\, Sel
 ami Ali Mah.\, Gazi Cad. No:22\, 34664 Üsküdar\, İstanbul
GEO:41.024366;29.033571
URL:https://latentshift.ai/schedule/rlaif-ile-model-finetuning/
CATEGORIES:LatentShift,Long talk
STATUS:CONFIRMED
TRANSP:OPAQUE
END:VEVENT
BEGIN:VEVENT
UID:ed01-run-the-dark-code@latentshift.ai
DTSTAMP:20261007T125617Z
DTSTART:20261017T090500Z
DTEND:20261017T092500Z
SUMMARY:Run the Dark Code: Why Your Agent's Error Handling Has Never Execut
 ed - and How to Run It on Purpose · Cenk Kalpakoglu
DESCRIPTION:Short talk · 20 min · English\n\nCenk Kalpakoglu · Vice Pres
 ident\, AI R&D\, Invicti\n\nYour agent chains LLM calls\, tool subprocesse
 s\, RAG lookups\, and HTTP requests. You wrote error handling for every li
 nk in that chain: the retry after a 429\, the cleanup when a tool gets kil
 led\, the fallback when your vector DB returns garbage. This talk's thesis
  is uncomfortably simple: you wrote that code\, but you've never seen it r
 un\, and none of your current testing tools can run it.\n\nWhy? Because ag
 ent failures don't live at a single layer. An SSE response that cuts off m
 id-token lives in HTTP semantics. A tool subprocess that gets OOM-killed l
 ives at the stdio layer. An ENOSPC lives at the syscall layer. Unit tests 
 test the mock. Kubernetes-level chaos tools inject at the wrong layer enti
 rely. Agent failures live at the SDK/API boundary\, and that's exactly whe
 re fault injection has to happen.\n\nI'll demonstrate this live with two m
 echanisms: an HTTPS proxy with a per-run ephemeral CA that synthesizes ven
 dor-accurate 429s and stream truncations over a real TLS connection\, so t
 he SDK sees exactly what production would send it\, and eBPF kprobes that 
 rewrite syscall returns in the kernel\, so the target gets a real ECONNRES
 ET\, not a simulated one. The tool is open source (faultkit)\, but this is
  a methods talk\, not a tool pitch. In the live demo\, we inject the llm-a
 pi-degraded scenario into a real agent test suite and find two red tests t
 hat didn't exist yesterday: two production incidents that won't exist tomo
 rrow.\n\nBonus lesson: a resilience test that passes green without firing 
 a fault is more dangerous than one that fails red. That's the story of exi
 t code 3.\n\nOne memorable takeaway:\n\n"Your error paths are code too. Do
 n't let production be the first place they run! Run them on purpose\, toda
 y."\n\nhttps://latentshift.ai/schedule/run-the-dark-code/
LOCATION:Bağlarbaşı Kongre ve Kültür Merkezi\, Boğaziçi Salonu\, Sel
 ami Ali Mah.\, Gazi Cad. No:22\, 34664 Üsküdar\, İstanbul
GEO:41.024366;29.033571
URL:https://latentshift.ai/schedule/run-the-dark-code/
CATEGORIES:LatentShift,Short talk
STATUS:CONFIRMED
TRANSP:OPAQUE
END:VEVENT
BEGIN:VEVENT
UID:ed01-kritik-sistemlerde-ai-modeli-gelistirmek@latentshift.ai
DTSTAMP:20261007T125617Z
DTSTART:20261017T093000Z
DTEND:20261017T095000Z
SUMMARY:Developing and Running AI Models in Critical Systems: How Do We Des
 ign Explainable\, Auditable\, and Reliable Models? · Fatih Bildirici
DESCRIPTION:Short talk · 20 min · Turkish\n\nFatih Bildirici · Senior Su
 bject Matter Expert\, ASELSAN\n\nTurkish Sign Language (TİD) interpretati
 on.\n\nThis talk is built around a fundamental lesson learned while deploy
 ing an AI model to production in a critical business process: a model prod
 ucing accurate predictions does not mean the system operates reliably. In 
 mission-critical / life-critical fields such as defense\, healthcare\, avi
 ation\, energy\, transportation\, and industrial control\, a model's outpu
 t becomes more than a score or a classification result. It becomes an oper
 ational signal that influences an alarm\, a human decision\, a procedure\,
  a risk level\, or fallback behavior.\n\nWithout sharing confidential deta
 ils about organizations\, data\, or systems\, the talk will describe the t
 echnical failure modes that can emerge when an AI model moves from a test 
 environment into a production decision chain. Initially\, the model's offl
 ine metrics appear acceptable: accuracy\, precision/recall\, latency\, and
  service response times meet their targets. But when real use begins\, the
  problem extends beyond model performance. The data distribution changes\,
  confidence scores become miscalibrated for some input segments\, explanat
 ions begin to differ for similar cases\, human intervention increases\, an
 d the system continues to process low-confidence outputs as candidates for
  decisions.\n\nBased on this experience\, the technical backbone of our ta
 lk will follow this flow:\n\ndata validation → feature pipeline → mode
 l inference → confidence / uncertainty estimation → explanation layer 
 → context check → risk gate → human review / fallback → monitoring
  → audit trail\n\nWithin this flow\, explainability will be treated as a
  control layer that conveys the model's limits to the system\, rather than
  a UI panel added later. An explanation will do more than answer "why did 
 the model produce this result?" It will also support the following signals
 : does the model recognize this data regime\, is its confidence truly reli
 able\, is the explanation stable\, is there drift\, are humans overriding 
 this output\, and when should the system give up on automated decision-mak
 ing?\n\nThe talk will examine these production lessons in technical detail
 : high confidence does not imply operational safety\, offline success and 
 production stability are different things\, explanation stability must be 
 monitored\, human-in-the-loop design cannot be just an approval screen\, a
 nd fallback/abstention mechanisms must be designed within the production d
 ecision chain\, rather than outside the model architecture.\n\nThe signals
  to monitor will go beyond accuracy\, precision\, recall\, p95 latency\, a
 nd cost/request. The talk will show why metrics such as expected calibrati
 on error\, confidence drift\, uncertainty distribution\, explanation drift
 \, explanation stability\, human override rate\, abstention rate\, fallbac
 k activation rate\, regression rate\, and audit completeness must also be 
 included.\n\nThe main lesson of the talk will be this: reliable AI in crit
 ical systems is more than a model that makes good predictions. Reliable AI
  is a system that knows when it is uncertain\, can explain its decisions\,
  controls risk\, hands over to a human when needed\, stops when needed\, a
 nd leaves a trace of every decision.\n\nhttps://latentshift.ai/schedule/kr
 itik-sistemlerde-ai-modeli-gelistirmek/
LOCATION:Bağlarbaşı Kongre ve Kültür Merkezi\, Boğaziçi Salonu\, Sel
 ami Ali Mah.\, Gazi Cad. No:22\, 34664 Üsküdar\, İstanbul
GEO:41.024366;29.033571
URL:https://latentshift.ai/schedule/kritik-sistemlerde-ai-modeli-gelistirme
 k/
CATEGORIES:LatentShift,Short talk
STATUS:CONFIRMED
TRANSP:OPAQUE
END:VEVENT
BEGIN:VEVENT
UID:ed01-lunch-1255@latentshift.ai
DTSTAMP:20261007T125617Z
DTSTART:20261017T095500Z
DTEND:20261017T103000Z
SUMMARY:Lunch · Omarchy 101 - Beautiful\, fun & agentic Linux for develope
 rs
DESCRIPTION:On stage during lunch: Omarchy 101 - Beautiful\, fun & agentic 
 Linux for developers\n\nhttps://latentshift.ai/schedule/
LOCATION:Bağlarbaşı Kongre ve Kültür Merkezi\, Boğaziçi Salonu\, Sel
 ami Ali Mah.\, Gazi Cad. No:22\, 34664 Üsküdar\, İstanbul
GEO:41.024366;29.033571
URL:https://latentshift.ai/schedule/
CATEGORIES:LatentShift
STATUS:CONFIRMED
TRANSP:OPAQUE
END:VEVENT
BEGIN:VEVENT
UID:ed01-demodan-operasyona-agent-surecleri@latentshift.ai
DTSTAMP:20261007T125617Z
DTSTART:20261017T103000Z
DTEND:20261017T111000Z
SUMMARY:Agent Workflows in Banking: From Demo to Operations · Şeyma Sarı
 gil
DESCRIPTION:Long talk · 40 min · Turkish\n\nŞeyma Sarıgil · Generative
  AI Architect\, Softtech\n\nTurkish Sign Language (TİD) interpretation.\n
 \nWhen we began our first experiments with generative AI agents for Türki
 ye İş Bankası in 2023\, our main question was quite simple: "How can we
  get AI to do this work?"\n\nAs we developed new agents for different busi
 ness areas\, our questions began to change. Was it enough for an agent to 
 produce the right answer? What would happen when the same system ran hundr
 eds or thousands of times in production? How would we detect incorrect dec
 isions? How would we know whether changes to prompts\, models\, or tools h
 ad broken existing workflows? When an agent failed\, how would we determin
 e whether the problem came from the model\, the data\, the tool\, or the o
 rchestration?\n\nOver time\, we saw that the real problem was keeping agen
 ts running safely\, observably\, and sustainably\, rather than simply deve
 loping them.\n\nIn this talk\, I will describe the real journey of the age
 ntic AI systems we developed in a banking environment from 2023 through la
 te 2026. I will share the approaches we tried as we moved from initial pro
 totypes to production workflows\, the problems we had not anticipated\, th
 e designs that failed\, and how these experiences led us to build a shared
  agent platform.\n\nThroughout the talk\, I will explore the following tra
 nsformation through real examples:\n\nMoving from developing individual ag
 ents to shared lifecycle management\n\nThe difference between saying "it w
 orks" and being truly reliable\n\nHow the need for evaluation\, observabil
 ity\, and AgentOps emerged\n\nManaging tool\, model\, prompt\, and agent v
 ersions together\n\nWhy human approval\, authorization\, and auditability 
 are part of the architecture in banking\n\nHow the duplication that arose 
 as each team developed its own agents created the need for a platform\n\nW
 hy debugging deep agents and multi-step workflows becomes increasingly dif
 ficult\n\nThis is not a story about "how we built the perfect architecture
 ." I will also openly share which assumptions proved wrong\, which solutio
 ns we had to redesign\, and which problems we are still trying to solve to
 day.\n\nhttps://latentshift.ai/schedule/demodan-operasyona-agent-surecleri
 /
LOCATION:Bağlarbaşı Kongre ve Kültür Merkezi\, Boğaziçi Salonu\, Sel
 ami Ali Mah.\, Gazi Cad. No:22\, 34664 Üsküdar\, İstanbul
GEO:41.024366;29.033571
URL:https://latentshift.ai/schedule/demodan-operasyona-agent-surecleri/
CATEGORIES:LatentShift,Long talk
STATUS:CONFIRMED
TRANSP:OPAQUE
END:VEVENT
BEGIN:VEVENT
UID:ed01-tabibert@latentshift.ai
DTSTAMP:20261007T125617Z
DTSTART:20261017T111500Z
DTEND:20261017T113500Z
SUMMARY:TabiBERT: A Large-Scale ModernBERT Foundation Model for Turkish · 
 Ebrar Kızıloğlu
DESCRIPTION:Short talk · 20 min · Turkish\n\nEbrar Kızıloğlu · Resear
 ch Assistant\, TABILAB\, Boğaziçi University\n\nTurkish Sign Language (T
 İD) interpretation.\n\nWe introduce TabiBERT (https://huggingface.co/boun
 -tabilab/TabiBERT)\, a state-of-the-art monolingual Turkish encoder based 
 on the ModernBERT architecture. Previously\, the Turkish NLP landscape lac
 ked a monolingual encoder trained from scratch that systematically incorpo
 rates modern architectural advances\, such as rotary positional embeddings
  and FlashAttention-based attention implementations. To bridge this gap\, 
 we trained TabiBERT on one trillion tokens sampled from an 84.88B-token mu
 lti-domain corpus containing web text (73%)\, scientific publications (20%
 )\, source code (6%)\, and mathematical content (0.3%). TabiBERT supports 
 a context length of 8\,192 tokens\, which is 16x that of the original BERT
  models\, and delivers up to a 2.65x inference speedup with significantly 
 reduced GPU memory overhead. To support rigorous evaluation\, we concurren
 tly developed TabiBench (https://huggingface.co/collections/boun-tabilab/t
 abibench)\, a comprehensive benchmark with 28 datasets across eight task c
 ategories. TabiBERT achieves a GLUE-style macro-average score of 77.58\, o
 utperforming the previous de facto standard\, BERTurk\, by 1.62 absolute p
 oints and establishing state-of-the-art results on five out of eight task 
 categories\, including major gains in question answering (+9.55 points).\n
 \nOur development process revealed several critical bottlenecks spanning c
 omputational infrastructure\, data quality\, and evaluation methodology.\n
 \nIn terms of computational and replicability challenges\, our initial pre
 -training trials on 8×H100 SXM GPUs experienced massive throughput degrad
 ation\, achieving only 30% of the speed documented in the ModernBERT paper
 . Hardware tests on 8×A100 GPUs yielded nearly identical performance\, co
 nfirming that the bottleneck lay entirely in the data loader rather than i
 n the accelerator hardware. Consultation with the ModernBERT authors revea
 led that high-throughput training requires a comprehensive pre-tokenizatio
 n pipeline. This hidden prerequisite highlighted a widespread problem rega
 rding replicability in large-scale modeling. Driven by this experience\, w
 e placed emphasis on maintaining high-quality documentation for our own pr
 e-training and fine-tuning procedures to facilitate future research and ad
 option.\n\nRegarding data quality and benchmarking challenges\, we discove
 red during downstream evaluation that several commonly used fine-tuning da
 tasets in Turkish NLP research were completely unstandardized. Most critic
 ally\, the field suffered from the lack of standardized data splits\, maki
 ng a true "apples-to-apples" comparison across different models entirely i
 mpossible. Furthermore\, several existing fine-tuning datasets exhibited q
 uality issues\, including inconsistent annotations\, labeling ambiguities\
 , and insufficient sample sizes. This systemic lack of reliable benchmarks
  led us to develop TabiBench\, in which we enforced a strict manual filter
 ing process to isolate high-fidelity evaluation targets and established co
 ncrete\, standardized train-test-validation splits to ensure reproducible 
 performance tracking.\n\nFinally\, we faced a distinct methodological chal
 lenge during token classification experiments (NER and POS tagging)\, in w
 hich initial subword-level evaluation yielded highly distorted results. Be
 cause different tokenizers segment the same text into varying numbers of s
 ubwords\, this discrepancy made subword-level scores incomparable across m
 odels\; architectures with higher token fertility could artificially infla
 te their scores by trivially predicting more fragmented tokens per word. T
 o address this\, we modified our methodology to use word-level evaluation\
 , aggregating predictions at word boundaries before computing metrics. Thi
 s approach ensures a fair comparison across models regardless of tokenizat
 ion strategy. Ultimately\, our work demonstrates that building a regional 
 language foundation model requires equal dedication to advanced architectu
 res\, explicit preprocessing transparency\, and robust evaluation methodol
 ogies.\n\nhttps://latentshift.ai/schedule/tabibert/
LOCATION:Bağlarbaşı Kongre ve Kültür Merkezi\, Boğaziçi Salonu\, Sel
 ami Ali Mah.\, Gazi Cad. No:22\, 34664 Üsküdar\, İstanbul
GEO:41.024366;29.033571
URL:https://latentshift.ai/schedule/tabibert/
CATEGORIES:LatentShift,Short talk
STATUS:CONFIRMED
TRANSP:OPAQUE
END:VEVENT
BEGIN:VEVENT
UID:ed01-evaluating-genai-agents-in-production@latentshift.ai
DTSTAMP:20261007T125617Z
DTSTART:20261017T114000Z
DTEND:20261017T120000Z
SUMMARY:Evaluating GenAI Agents in Production: Lessons from Booking.com · 
 Ilkcan Keles
DESCRIPTION:Short talk · 20 min · English\n\nIlkcan Keles · Senior Machi
 ne Learning Engineer\, Booking.com\n\nShipping an LLM-based agent is the e
 asy part. Knowing whether it actually works once it is live - breaking dow
 n problems\, calling external tools\, and iterating through multi-step rea
 soning against real user traffic - is where most teams get stuck. Evaluati
 on that works for a single-model prompt falls apart the moment the system 
 can take actions\, and public benchmark scores tell you almost nothing abo
 ut how the agent behaves on your product. This is a talk about what we act
 ually do at Booking.com to evaluate AI agents in production\, with the rea
 l numbers and trade-offs behind the framework.\n\nWe start with the founda
 tion that everything else rests on: LLM-as-a-Judge. We cover how we build 
 golden datasets that represent the true production distribution - includin
 g the edge cases - with annotation protocols rigorous enough to trust\, an
 d how we develop judge-LLMs through iterative prompt engineering with chai
 n-of-thought reasoning\, then deploy them for continuous automated evaluat
 ion at scale. The practical detail matters here: a strong model to set the
  upper bound in development\, a cheaper model to run in production without
  breaking the cost budget.\n\nThe core of the talk is our dual approach to
  agent evaluation. Black box evaluation measures task completion - the fra
 ction of expected tasks correctly completed per conversation - using a jud
 ge LLM that receives the full dialogue\, the predefined task list\, and pr
 oduct-specific constraints. Glass box evaluation looks inside the agent to
  understand why it behaved the way it did. We assess tool proficiency thro
 ugh tool call validity (checking calls against JSON schemas for correct na
 mes\, required parameters\, types\, and value formats) and tool correctnes
 s (whether the agent picks the right tools\, including correctly declining
  when no relevant tool exists). We dig into tool argument hallucination - 
 the agent calls the right tool but invents wrong parameter values\, "Paris
 \, France" versus "Paris\, Texas" - and tool reliability: a set of judge-L
 LM checks ensuring tool names and descriptions are clear\, concise\, and m
 achine-interpretable\, because obscure or verbose specifications quietly d
 egrade agent performance.\n\nWe then get concrete about benchmarking. Befo
 re an agent earns its added complexity\, it has to beat simpler baselines 
 - zero-shot LLM\, few-shot LLM\, deterministic LLM chains - on performance
 \, cost\, and latency\, so the decision to deploy is data-driven rather th
 an hype-driven. Finally we walk through the frontier problems we are still
  wrestling with: consistency\, where pass^k across paraphrased queries dro
 ps below 20% after just 8 trials\; reasoning efficiency as task complexity
  grows\; memory quality and its measurable impact on task completion\; and
  trajectory optimality assessed with preference-based judges.\n\nAttendees
  leave with a concrete\, reusable playbook for evaluating AI agents in pro
 duction - the metrics\, the judge prompts\, the baselines\, and the failur
 e modes - grounded in real-world experience running agents at one of the w
 orld's largest travel platforms.\n\nhttps://latentshift.ai/schedule/evalua
 ting-genai-agents-in-production/
LOCATION:Bağlarbaşı Kongre ve Kültür Merkezi\, Boğaziçi Salonu\, Sel
 ami Ali Mah.\, Gazi Cad. No:22\, 34664 Üsküdar\, İstanbul
GEO:41.024366;29.033571
URL:https://latentshift.ai/schedule/evaluating-genai-agents-in-production/
CATEGORIES:LatentShift,Short talk
STATUS:CONFIRMED
TRANSP:OPAQUE
END:VEVENT
BEGIN:VEVENT
UID:ed01-coffee-1505@latentshift.ai
DTSTAMP:20261007T125617Z
DTSTART:20261017T120500Z
DTEND:20261017T123000Z
SUMMARY:Coffee break
LOCATION:Bağlarbaşı Kongre ve Kültür Merkezi\, Boğaziçi Salonu\, Sel
 ami Ali Mah.\, Gazi Cad. No:22\, 34664 Üsküdar\, İstanbul
GEO:41.024366;29.033571
URL:https://latentshift.ai/schedule/
CATEGORIES:LatentShift
STATUS:CONFIRMED
TRANSP:OPAQUE
END:VEVENT
BEGIN:VEVENT
UID:ed01-the-productivity-illusion@latentshift.ai
DTSTAMP:20261007T125617Z
DTSTART:20261017T123000Z
DTEND:20261017T131000Z
SUMMARY:The Productivity Illusion: What AI Really Changed for Developers ·
  Fatih Kadir Akın
DESCRIPTION:Long talk · 40 min · Turkish\n\nFatih Kadir Akın · Develope
 r Advocate\, WordPress\, Automattic\n\nTurkish Sign Language (TİD) interp
 retation.\n\nEveryone says AI made developers faster. But did it really? T
 his session dives into the hidden layers of AI-assisted development: How i
 t affects focus\, flow\, and trust. Drawing from real-world experience bui
 lding AI tools and observing developer behavior\, We'll reveal what we gai
 ned\, what we lost\, and what we must redesign to make productivity human 
 again.\n\nhttps://latentshift.ai/schedule/the-productivity-illusion/
LOCATION:Bağlarbaşı Kongre ve Kültür Merkezi\, Boğaziçi Salonu\, Sel
 ami Ali Mah.\, Gazi Cad. No:22\, 34664 Üsküdar\, İstanbul
GEO:41.024366;29.033571
URL:https://latentshift.ai/schedule/the-productivity-illusion/
CATEGORIES:LatentShift,Long talk
STATUS:CONFIRMED
TRANSP:OPAQUE
END:VEVENT
BEGIN:VEVENT
UID:ed01-fully-autonomous-pull-requests@latentshift.ai
DTSTAMP:20261007T125617Z
DTSTART:20261017T131500Z
DTEND:20261017T133500Z
SUMMARY:Fully Autonomous Pull Requests Didn't Lower the Bar · Kaya Gokalp
DESCRIPTION:Short talk · 20 min · English\n\nKaya Gokalp · Staff Softwar
 e Engineer\, Sonar\n\nLast month\, more than a third of the pull requests 
 merged through our platform had no human reviewer. Our agent was the only 
 thing that read them. About 8% of them were merged fully autonomously by u
 s after we approved them.\n\nWe can measure this because we ship the auto-
 approve and auto-merge paths\, which gives us unique data nobody else has.
  About 6k pull requests a week merged with our agent as the only reviewer.
  They merged in a median of 0.5 hours against roughly 2.5 hours for human-
 reviewed ones. They were also larger\, averaging about 2k changed lines ag
 ainst 1.5k\, so this is not teams skipping review on typo fixes.\n\nThe re
 view did not get worse when the agent was the only one looking. About 13% 
 of pull requests had changes requested in both groups. The agent blocked s
 lightly more often with no human present\, 2% against 1.7%. Developers dis
 missed its findings less often too\, 10% against 16%.\n\nhttps://latentshi
 ft.ai/schedule/fully-autonomous-pull-requests/
LOCATION:Bağlarbaşı Kongre ve Kültür Merkezi\, Boğaziçi Salonu\, Sel
 ami Ali Mah.\, Gazi Cad. No:22\, 34664 Üsküdar\, İstanbul
GEO:41.024366;29.033571
URL:https://latentshift.ai/schedule/fully-autonomous-pull-requests/
CATEGORIES:LatentShift,Short talk
STATUS:CONFIRMED
TRANSP:OPAQUE
END:VEVENT
BEGIN:VEVENT
UID:ed01-closing-1640@latentshift.ai
DTSTAMP:20261007T125617Z
DTSTART:20261017T134000Z
DTEND:20261017T135000Z
SUMMARY:Closing · The draw: ASUS Dual GeForce RTX 5060 Ti OC 8GB
DESCRIPTION:The draw: ASUS Dual GeForce RTX 5060 Ti OC 8GB. We draw one nam
 e on stage before the closing words. The RTX 5060 Ti goes home with them.\
 n\nhttps://latentshift.ai/schedule/
LOCATION:Bağlarbaşı Kongre ve Kültür Merkezi\, Boğaziçi Salonu\, Sel
 ami Ali Mah.\, Gazi Cad. No:22\, 34664 Üsküdar\, İstanbul
GEO:41.024366;29.033571
URL:https://latentshift.ai/schedule/
CATEGORIES:LatentShift
STATUS:CONFIRMED
TRANSP:OPAQUE
END:VEVENT
END:VCALENDAR