---
title: "Simulation-Maxxing: How Nubank ships agents 20× faster with simulations"
category: "talks"
date: "2026-07-01"
time: "2:50pm-3:10pm"
track: "AI in Finance"
room: "Track 3"
speakers: ["Shreya Rajpal", "Aman Gupta"]
sourceLabels: ["Official conference schedule", "Public YouTube metadata"]
scheduleTrack: "AI in Finance"
scheduleRoom: "Track 3"
scheduleLabels: ["AI in Finance", "Track 3", "session", "confirmed"]
---
# Simulation-Maxxing: How Nubank ships agents 20× faster with simulations

## Conference Context
- Date/time: 2026-07-01 · 2:50pm-3:10pm
- Track/room: AI in Finance · Track 3
- Speaker(s): Shreya Rajpal, Aman Gupta
- Session type/status: session · confirmed

- Track: AI in Finance
- Room: Track 3
- Session type: session
- Status: confirmed

## Session Description
You know how to build an agent - write a prompt, spec out some tools and call an LLM (or gateway). At this point, you probably also know how to build an agent that “actually works” using some combination of agent frameworks, eval tools and looking at your data. This talk is about building an agent much, much faster using simulations to hill-climb your agent configuration instead of grinding on real data. We’ll dive deep into a case study of how a top-5 fintech made their agent dev cycle 20x faster using simulation-driven optimization. We’ll cover: - When to use real data vs. simulations in agent building - How to design simulation environments tailored to your agent - How to automate the optimization loop so you’re hill climbing agent configurations without manual tuning

## Media Evidence
[Trust, but Verify: Shreya Rajpal](https://www.youtube.com/watch?v=9-vGxMoUM9Y) (speaker-match related prior/adjacent AI Engineer video; captions: English auto-captions).

- Source video: `youtube-9-vGxMoUM9Y`
- Slide deck: [[youtube-9-vGxMoUM9Y-dense-slides|Dense Slides: Trust, but Verify: Shreya Rajpal]] — 13 visible slide image(s); 13 HTML recreation(s).
![[assets/dense-slides/9-vGxMoUM9Y/slide-001.jpg]]
![[assets/dense-slides/9-vGxMoUM9Y/slide-002.jpg]]
![[assets/dense-slides/9-vGxMoUM9Y/slide-003.jpg]]
- Additional slide evidence: [[youtube-9-vGxMoUM9Y-slides|Slides: Trust, but Verify: Shreya Rajpal]], [[youtube-9-vGxMoUM9Y-reconstructed-slides|Reconstructed Slides: Trust, but Verify: Shreya Rajpal]]
- Slide-derived themes for `youtube-9-vGxMoUM9Y`: current, guardrails, self, driving, cars, classical, deep, learning.

## Evidence Graph
This evidence graph is generated from currently linked source material: official schedule text, related video pages, cached transcripts, visible slide text, dense/reconstructed slide pages, and AI slide-classification audits.

### Media Signals
- `youtube-9-vGxMoUM9Y` — 9 slide-derived text signals
- Slide-derived themes for `youtube-9-vGxMoUM9Y`: current, guardrails, self, driving, cars, classical, deep, learning.
- Evidence links for `youtube-9-vGxMoUM9Y`: [[youtube-9-vGxMoUM9Y]], [[youtube-9-vGxMoUM9Y-slides]], [[youtube-9-vGxMoUM9Y-dense-slides]], [[youtube-9-vGxMoUM9Y-reconstructed-slides]]

### Agent Reading Notes
Use these signals to refine the synopsis, topic links, people/company context, and method notes. If a source is a related external video rather than an exact official recording, keep it framed as supporting evidence.

## Transcript Status
Related video transcript availability: English auto-captions. Treat this as supporting context, not a recording of this exact scheduled session unless later confirmed. Not fetched yet.

## People
- [[shreya-rajpal]]
- [[aman-gupta]]

## Supporting Slides
- [[youtube-9-vGxMoUM9Y-slides]] — extracted from the related public AI Engineer video.

## Slide Evidence
- Slide-only cropped deck: [[youtube-9-vGxMoUM9Y-dense-slides]] (13 viable slide images).
- Related slide/OCR pages:
- [[youtube-9-vGxMoUM9Y-dense-slides]]
- [[youtube-9-vGxMoUM9Y-reconstructed-slides]]
- [[youtube-9-vGxMoUM9Y-slides]]
- Slide-derived terms: `guardrails`, `source`, `change`, `llms`, `library`, `validators`, `self`, `deep`, `querdraie`, `current`, `cofounder`, `past`, `infra`, `lead`, `mlops`, `driving`, `cars`, `classical`

## Synthesis
### Synthesized Breakdown
# Simulation-Maxxing: How Nubank ships agents 20× faster with simulations ## Conference Context - Date/time: 2026-07-01 · 2:50pm-3:10pm - Track/room: AI in Finance · Track 3 - Speaker(s): Shreya Rajpal, Aman Gupta - Session type/status: session · confirmed - Track: AI in Finance - Room: Track 3 - Session type: session - Status: confirmed ## Session Description You know how to build an agent - write a prompt, spec out some tools and call an LLM (or gateway). At this point, you probably also know how to build an agent that “actually works” using some combination of agent frameworks, eval tools and looking at your data. This talk is about building an agent much, much faster using simulations to hill-climb your agent configuration instead of grinding on real data. We’ll dive deep into a case study of how a top-5 fintech made their agent dev cycle 20x faster using simulation-driven optimization.

### Speaker And Company Context
- [[shreya-rajpal|Shreya Rajpal]] — CEO at [[snowglobe|Snowglobe]].
- [[aman-gupta|Aman Gupta]] — Principal Machine Learning Engineer at [[nubank|Nubank]].

### Topics Covered
- [[agent-security]]

### Derived Links And Source Material
- [[youtube-9-vGxMoUM9Y]] — related YouTube source page.
- [[youtube-9-vGxMoUM9Y-slides]] — slide evidence.
- [[youtube-9-vGxMoUM9Y-reconstructed-slides]] — slide evidence.
- [[youtube-9-vGxMoUM9Y-dense-slides]] — slide evidence.

### Novel Concepts / Clever Methods
- No highlighted novel concept has been detected yet.

### Evidence Boundary
This synthesis is based on the official schedule and linked source pages. It should be revisited when exact session recordings or transcript-backed secondary sources are available.