# QCon San Francisco 2025: 10 Standout AI Sessions on LLMs, Agents, and GenAI Infrastructure

> QCon SF 2025 offers 10 standout AI sessions for engineering leaders, from LLM UX to multi-model serving. Learn patterns to ship faster, cut risk, and show ROI, Nov 17-21.

- Type: AI news
- Page: https://completeaitraining.com/news/qcon-san-francisco-2025-10-standout-ai-sessions-on-llms/
- Published: 2025-10-15
- Updated: 2025-10-15
- Author: Jeroen Erne
- Jobs: IT and development
- Reading time: 3 min

### 10 AI-Related Standout Sessions at QCon San Francisco 2025

QCon San Francisco 2025 lands November 17-21 at the Hyatt Regency San Francisco. Across 15 tracks, AI threads through almost everything software leaders care about: speed, safety, scale, and real results. Here are 10 sessions (in no particular order) worth your attention if you build or lead engineering teams.

#### "Accelerating LLM-Driven Developer Productivity at Zoox" - Amit Navindgi @Zoox

A practical blueprint for scaling an organization's AI capabilities with real design patterns and org strategies. Expect tactics you can apply without boiling the ocean.

- Templates for internal copilots, review bots, and automated QA
- Governance patterns for prompts, evals, and data boundaries
- Metrics that track shipped features, cycle time, and model ROI

#### "Engineering at AI Speed: Lessons from the First Agentically Accelerated Software Project" - Adam Wolff @Anthropic

How Claude Code's architecture favors speed over complexity, with agents that help but don't block. Useful for teams balancing experimentation with production-grade reliability.

- Thin-agent patterns that keep humans in control
- Event-driven flows, retries, and deterministic fallbacks
- What to simplify first to reduce latency and surprises

#### "Deep Research for Enterprise: Unlocking Actionable Intelligence from Complex Enterprise Data with Agentic AI" - Vinaya Polamreddi @Glean

Turning messy, multi-system enterprise data into answers people trust. Covers scalable design and advanced training methods for agentic research workflows.

- Cross-source retrieval with strict permissioning
- Grounding, provenance, and audit trails for compliance
- E2E evaluation: precision, freshness, and task completion

#### "The Future of Engineering: Mindsets That Matter When Code Isn't Enough" - Ben Greene @Tessi

The engineer's role is expanding beyond code. This session reframes impact across systems, products, and human workflows.

- Systems thinking over ticket thinking
- Prompt + product + process as a single toolchain
- Outcome metrics: time saved, error reduction, enablement

#### "Dynamic Moments: Weaving LLMs into Deep Personalization at DoorDash" - Sudeep Das and Pradeep Muthukrishnan @DoorDash

How to blend LLMs with real-time signals to personalize at scale. Strong fit for teams owning growth, recommendations, and marketplace dynamics.

- Feature pipelines that respond in milliseconds
- Safety rails for toxic content and bias
- Online testing strategies: explore/exploit without regressions

#### "Designing Fast, Delightful UX with LLMs in Mobile Frontends" - Bala Ramdoss @Amazon

UX principles that make LLM features feel instant and reliable on mobile. Architecture meets design to remove friction.

- Latency budgets, streaming UI, and optimistic updates
- Fallbacks, timeouts, and partial responses that don't frustrate
- Offline strategies and smart caching for flaky networks

#### "One Platform to Serve Them All: Autoscaling Multi-Model LLM Serving" - Meryem Arik @Doubleword

A technical breakdown of multi-model serving without runaway costs. Covers shared base weights, hot-swapped adapters, and smart eviction.

- Adapter-based routing (LoRA, prefix-tuning) with fast swaps
- Dynamic loading and tiered memory for GPU efficiency
- Autoscaling on tokens/sec, not just requests/sec

#### "From Content to Agents: Scaling LLM Post-Training Through Real-World Applications and Simulation" - Faye Zhang @Pinterest and Andi Partov @Veris AI

Post-training that moves beyond benchmarks into production. From content generation to simulation-based agent training that learns from real tasks.

- Leveraging production data for SFT and preference tuning
- Offline RLHF and simulator-driven evaluation loops
- Agent tooling, planning, and action constraints

#### "Powering the Future: Building Your GenAI Infrastructure Stack" - Maggie Hu and Merrin Kurian @Intuit

Inside a platform that serves ~100M users with vector stores, prompt management, RAG pipelines, and agent orchestration.

- Reference stack: data layer, retrieval, orchestration, observability
- Multi-tenant security, PII handling, and policy enforcement
- Cost controls: caching, compression, model routing

#### "AI-Driven Productivity: From Idea to Impact" - Jyothi Nookula @Stealth Startup

A pragmatic framework to convert GenAI enthusiasm into shipped product and measurable impact. Less slideware, more scorecards.

- Idea triage using value vs. feasibility scoring
- Proof-of-value before proof-of-concept
- Change management: training, playbooks, and adoption metrics

### Why this matters

These sessions focus on what actually ships: lower latency, safer outputs, repeatable infrastructure, and clear ROI. If you lead teams, you'll find patterns to reduce risk while moving fast.

### Attend and go deeper

QCon San Francisco 2025 runs November 17-21 at the Hyatt Regency San Francisco. Check the full schedule and details on the official site.

[QCon San Francisco 2025 - Official Site](https://qconsf.com/?ref=completeaitraining.com)

If you're building a training plan for engineers and product teams, explore curated AI learning paths and certifications:

- [AI courses by job role](https://completeaitraining.com/courses-by-job/)
- [Popular AI certifications](https://completeaitraining.com/popular-certifications/)

---

This is the Markdown form of https://completeaitraining.com/news/qcon-san-francisco-2025-10-standout-ai-sessions-on-llms/. The whole site is mapped at https://completeaitraining.com/llms.txt.
