Google Developer Group - Chapel Hill

• Step 1: Click the RSVP button on this event page to register (it's free). IF YOU ARE READING THIS ON MEETUP.COM, you MUST RSVP on our GDG event platform to attend: https://gdg.community.dev/events/details/google-gdg-chapel-hill-presents-applied-ai-conference-2026/
• Step 2: Visit the event page on July 18, where a "Join" button will appear when the event starts.
• Step 3: Browse the tentative schedule at gdg-chapel-hill.github.io/applied-ai-conference-2026. (We recommend bookmarking it! It's more readable than the agenda below).

### Agenda · Saturday, July 18, 2026 (ALL TIMES MENTIONED BELOW ARE EDT)

• 8:30 - 9:00 AM · Conference orientation
• 9:00 – 9:30 AM · I Don't Trust AI Agents (And Neither Should You): Building Production-Ready Architectures
  Darko Mesaroš · Principal Developer Advocate @ AWS · Technical · 30 min
  Your AI agent works great in the demo. Then you deploy it and it hallucinates a refund policy, exposes customer data, or just loops endlessly burning tokens. This session walks through a layered approach to agent safety using Amazon Bedrock AgentCore and the Strands Agents SDK: guardrails, observability, multi-agent safety patterns, and reference architectures you can adapt immediately.
• 9:45 – 10:15 AM · Shipping Safe AI Agents: A Production Safety Playbook from 2,500 Deployments
  Lorenzo Satta Chiris · Director of Excode · Technical · 30 min
  A condensed safety playbook drawn from 2,500 real deployments and co-authoring the AURA open-source agent risk framework. Three real failure modes: an agent that escalated its own permissions, a RAG pipeline that confidently cited fake policy docs, and multi-agent outputs that contradicted each other on a customer-facing response, with the root cause, the fix, and the monitoring signal that now catches each one early.
• 10:30 – 11:15 AM · From Vibes to Verified: Building an Autonomous Eval + Fix Engine for AI Agents
  Shashank Agarwal · Founder & CEO, Noveum.ai · Ex-AWS SageMaker (2nd Engineer) · Technical · 45 min
  Gartner predicts 40% of agentic AI projects will be cancelled by 2027, mostly because teams are shipping on "vibes-based engineering." This session walks through an autonomous eval and remediation engine that runs 106+ scorers across 18 categories, and delivers fixes as actual merged pull requests, not suggestions. Real enterprise results: 200x faster time-to-fix, 4–6x performance improvement across 9 production deployments.
• 11:30 – 11:45 AM · AI Can Write Code. It Can't See the System.
  Eshwar Yaddanapudi · Creator of Fluent-Graph · Ex-EM, ServiceNow · ⚡ Lightning · 15 min
  AI generates code fast, but it can't see how that code interacts with everything else. A single AI-assisted change can propagate failures through hidden dependencies undetected until after deployment. This lightning talk presents one idea: move dependency visibility to the point of change using system-level dependency graphs, shifting from reactive observability to pre-execution reasoning.
• 12:00 – 1:00 PM · 🍽 Lunch break
• 1:15 – 1:45 PM · When Agents Break: Designing Fault-Tolerant Multi-Agent Orchestration Beyond LangChain
  Ravi Kiran Pagidi · Senior AI Data Engineer, Navy Federal Credit Union · Technical · 30 min
  A five-agent LangChain system worked in staging and fell apart in production within the first week, with cascading breakdowns where one bad link silently poisoned the entire chain. This talk walks through the custom orchestration layer built on top of LangChain: agent-level circuit breakers, context validation on every handoff, fallback routing, and the monitoring dashboards that catch silent failures. Patterns applicable to LangChain, CrewAI, AutoGen, or custom stacks.
• 2:00 – 2:15 PM · Using Claude's web_search Tool to Track AI Citations: An Open-Source Python Pattern
  Ignacio Lopez · Fractional Head of AI, Work-Smart.ai · Bilingual EN/ES · ⚡ Lightning · 15 min
  "AI visibility" SaaS tools run $300–$500/month per site. This talk walks through a free MIT-licensed alternative built as four Python scripts, using Claude's web_search tool to ask buyer-intent questions and parse which domains get cited. Costs about $1–$3 per run via the Anthropic API. Includes the exact API call structure, parsing approach, and the five query patterns that surface real citation behavior.
• 2:30 – 3:00 PM · Agent Memory: Building Stateful AI Agents That Remember, Adapt, and Work Across Time
  Ben Labaschin · Principal AI/ML Engineer @ Workhelix · O'Reilly Author · Technical · 30 min
  Most teams treat agent memory as a storage problem: add a vector database, keep more context, expect improvement. In production, the harder problem is what deserves to become memory, how to retrieve the right memory at the right time, and how to prevent old memories from quietly degrading the system. Drawing on lessons from production systems and a forthcoming O'Reilly book, this talk walks through the core design decisions: what to store, what to forget, how to scope retrieval, and how to maintain memory as it ages.
• 3:15 – 4:00 PM · UISurf: Toward Universal UI Automation with Cross-Environment Agents
  Henry Ruiz · Research Scientist @ Texas A&M AgriLife Research · GDE in AI & Google Cloud · 🔧 Demo · 45 min
  UISurf is an open-source multimodal agentic UI automation platform that enables AI agents to perceive, reason, and collaborate across browser and desktop environments to complete end-to-end tasks. The demo covers its multi-agent architecture: planning, browser, desktop, automation, and summarization agents coordinated through Agent-to-Agent (A2A) communication, supporting both fully autonomous and human-in-the-loop execution modes.
• 4:15 – 4:45 PM · 10x Engineering: Battle-Tested Lessons from Transforming Teams with GenAI
  Dennis Nerush · Director of AI Engineering @ Elementor · Technical · 30 min
  What happens when generative AI becomes the backbone of how an engineering team actually works? Over the past year, this team multiplied productivity, cut delivery times, and unlocked new collaborative workflows that spread across the company. Practical lessons and battle-tested insights: what worked brilliantly, what failed and why, and the cultural shifts needed to make GenAI a force multiplier rather than just another tool.
• 5:00 – 5:30 PM · From Prompts to Design-Driven Specs: Lessons in Taming AI Chaos
  Steve Fox · Founder, cleanapi.ai · Technical · 30 min
  Spec-driven AI codegen promises the spec as the durable source of truth, but in practice the code drifts on every iteration until the spec becomes just documentation. This case study walks through three iterations of trying to make the spec actually durable: the constraint each iteration revealed, and the system that finally worked. Closes with a live demo. You'll leave with three principles earned the hard way.
• 5:45 – 6:15 PM · Spec-Driven Agent Development with ADK & Antigravity
  Jitendra Gupta · Enterprise Architect, Cloud & AI @ EPAM Systems · 🔧 Demo · 30 min
  A hands-on walkthrough of Spec-Driven Development (SDD) using Google's Agent Development Kit (ADK) and the Antigravity agent-first IDE. Move beyond prompt-based coding to a structured, artifact-first approach where specifications, plans, and tasks drive implementation, walking through the full SDD lifecycle: specify, clarify, plan, analyze, and implement.