Accepting submissions till 31 Oct 2026, 11:59 PM
Not accepting submissions
Submission deadline - 31 October 2026
Make a submission
Accepting submissions till 31 Oct 2026, 11:59 PM
|
KB
Karan Bansal  Hardening AI Coding Agents with Hooks: Enforcing Least Privilege on Autonomous DevelopersAbstract Teams are putting AI coding agents into production workflows faster than they are putting controls around them. An agent that writes code also runs shell commands, reads secrets, installs packages, and pushes to git, at machine speed. This talk is about the control layer that has held up in production: event-driven hooks that intercept every tool call and allow, deny, or escalate it befo… more
Type of submission: 30 mins talk
|
Tushar Shah Six Agents, One Reality Check: What Actually Made Our Production Agent ReliableAbstract We often treat multi-agent architecture as the solution to complex AI workflows. In production, the harder work is building the harness around the model: clear agent boundaries, useful retrieval, safe tool access, human-review points, and evaluation that measures real outcomes. In this talk, I will share lessons from a six-agent LangGraph workflow for prospect research and outbound opera… more
Type of submission: 30 mins talk
|
Khaja Shaik  Evidence-Governed Agent Harnesses (EGAH)Evidence as first-class runtime state in long-running agent workflows. more
Type of submission: 30 mins talk
|
Swetha A When Good Agents Disagreed Until We Built Shared UnderstandingAbstract Multi-agent systems are often built around orchestration: an orchestrator routes requests to specialist agents, coordinates their work, and brings their outputs together. But in production, we encountered a more fundamental problem: every agent could be doing its job correctly, while the system as a whole was still wrong. more
Type of submission: 30 mins talk
|
|
Swetha A Understanding Your Agents From Traces to TrustAbstract While building and operating LLM-powered systems in production, one lesson keeps repeating itself which is that the hardest failures are often invisible. Unlike traditional software, agentic systems can fail in countless ways while every dashboard remains green. An agent may misunderstand user intent, choose the wrong tool, retrieve stale information, drift from its objective, or silentl… more
Type of submission: 30 mins talk
|
|
UK
Utkarsh Kanwat Your Agent Can Run as Long as You Can Check ItCoding agents can now work for hours without help, but in most other kinds of work an agent still needs a person to step in every so often. It’s the same models in both cases, so the model isn’t what decides how long an agent can run. The checks around it are. Every step an agent takes builds on its last output, and small mistakes pile up unless something outside the agent catches them. The agent… more
Type of submission: 30 mins talk
|
|
Ar
Akash raj Agents with no internet: building a document intelligence platform for an air-gapped defence settingAbout the session Most agent stacks quietly assume the internet: hosted models, cloud vector stores, external tools and telemetry. For a defence organisation (anonymised), none of that was allowed. The system had to run fully air-gapped, with AI agents on a single machine, an Indian-origin language model hosted on-premise, and no data leaving the room. The job was to help engineers review large v… more
Type of submission: 30 mins talk
|
Mahita D The Hard Part Isn't Building an Agent. It's Changing One That's Already in Production.What happens when the underlying API that your agentic system relies on changes unexpectedly? This talk is a production case study of how I handled an upstream API change in two weeks for a system already running in production, without regressions. more
Type of submission: 30 mins talk
|
|
AB
Ashok Bugude After the demo: wiring a voice agent into the CRM without breaking eitherA voice assistant that explains a telecom bill looks finished in a demo. The moment it has to identify the caller, read billing data and take an action through a CRM API, it becomes a different system: every conversational turn has a latency budget, the CRM’s API contract drifts, calls fail halfway through, and some actions should never be taken without a human. This talk walks through the integr… more
Type of submission: 30 mins talk
|
|
SP
Srinivas Padmanabhuni Is your LLM eval measuring the right thing?Format: 30-minute talk (no demo) The problem: Teams ship LLM, RAG and agent systems on eval scores that often measure the wrong thing. Judge models drift and disagree, generic metrics miss faithfulness and safety failures, and eval suites stop reflecting production once agents start calling tools. more
Type of submission: 30 mins talk
|
|
KR
Kumar Ravi Our LLM Agents Passed the Evals. Then They Met Real Users.Preferred format: 30 minutes, talk without demo. Topic: Agents in production. more
Type of submission: 30 mins talk
|
|
DG
Divyansh Golyan A hundred agents in 7.4 MBFormat: 15-minute flash talk with a live demo, or 45 minutes if you’d like the full version. more
Type of submission: Demo - audiences open laptops and try the demo
|
|
AD
Archit Dhamija When memory goes stale: what 367 commits did to our coding agents' knowledgeEvery engineer on a team now runs a coding agent, and every agent builds its own picture of the codebase. The shared layer on top - decisions, conventions, the rules in CLAUDE.md or AGENTS.md - goes stale the moment code changes under it, and an agent acts on an outdated fact with the same confidence as a true one. more
Type of submission: 30 mins talk
|
|
SK
Shivam Khandelwal [Add a catchy title]{Describe your session in 2 paragraphs} {Mention 1-2 takeaways from your session} more
Type of submission: 30 mins talk
|
|
JP
Jyothi Pujari Seven Ways Our Own Benchmarks Lied to Us, and the Preflight That Catches ThemThe problem. Before launching a multi-tenant voice agent, we had to decide how many phone calls one GPU could carry for a self-hosted model we are evaluating (Gemma 4 26B-A4B on an RTX PRO 6000). Our first load test said 16 calls at about 0.3 s to the first token. It was wrong: every stream sent the same prompt, so the server cached one copy. With a different prompt per call, as different tenants… more
Type of submission: 30 mins talk
|