---
title: "Slides: Trust, but Verify: Shreya Rajpal"
category: "slides"
video_id: "9-vGxMoUM9Y"
sourceLabels: ["Public YouTube video frames", "Public YouTube metadata"]
---

# Slides: Trust, but Verify: Shreya Rajpal

## Source Video
[Trust, but Verify: Shreya Rajpal](https://www.youtube.com/watch?v=9-vGxMoUM9Y)

## Relationship To World's Fair 2026
These slides are extracted from a public AI Engineer YouTube video connected to World's Fair 2026. Speaker-matched clips are supporting context unless later confirmed as exact session recordings; official livestream recordings are day-level/event-level source material.

## Related Scheduled Sessions
- No individual scheduled session mapping has been assigned yet; treat this as an event livestream deck.

## Extracted Slides
![[assets/slides/9-vGxMoUM9Y/slide-001.jpg]]

OCR text:

> About me
> r Current CEO & Cofounder @ Guardrails Al
> » ‘ Past ML infra lead @ MLops Co,
> —— F ML @ Self driving cars,
> . “ZB ‘ Classical Al & Deep learning research
> x > Gears
> a uniteges «ORE Tome gw pawns ny i « woe
> + PF ———>+ + > .
> —_—— i rT
> —— oer a“ acre BONE Bons BD,
> 
> ri °
> AuToget, = GC

![[assets/slides/9-vGxMoUM9Y/slide-002.jpg]]

OCR text:

> About me
> a Current CEO & Cofounder @ Guardrails Al
> > - Past ML Infra lead @ MLops Co,
> sort ML @ Self driving cars,
> . wr \ Classical Al & Deep learning research
> ~ r :
> S . Guacdraiis
> | — AI
> Reseaurcn A gue cations ~ . AL. wn el !
> niclassen AL Deep Cearrng Mi Fee Se deny cart, & MLOE Co '
> + ——___——> <« ——___> , +—— > .
> TOT ih
> :« aaririn 2014 2016 2021 2023

![[assets/slides/9-vGxMoUM9Y/slide-003.jpg]]

OCR text:

> We're seeing a cambrian explosion of applications in Al
> Auto~ GPT May Be The Seriously, software engineering as we have known it is dead in the water.
> Strong AI Tool That oe . ;
> Surpasses ChatGPT This Al is better than at least 95% of coders. And to the extent it is not, it
> 
> delivers 1000x productivity improvements to complement them.
> We have now officially entered the Age of Al.
> Ge eAG REE ahem eae
> \ ce vr fees tecy Mew Generative Al Will Change tales
> Setee
> How Generative Al Will
> + Quarctrate Al Change Sales
> 
> by Prabhiskarn Sinha. Arun Shastn, and Satty £. Lonerae

![[assets/slides/9-vGxMoUM9Y/slide-004.jpg]]

OCR text:

> The reality One Month Retention
> incumbents
> es =~
> ee
> Path to 100 Million Users ..... ee ~*~
> 
> es °~
> 
> ~~ ~~ -_ FS rrrrrrr——sY
> 
> © ¢ ® @ —————
> , / eS --~
> es «~~
> Z -
> . 63% Mechen
> oe Al-First Companies
> ~~ sere nominees ee
> | rc tC
> ee >
> a
> 42% Mechan
> ‘+ Querdraie Al
> Source: https://www.sequoiacap.com/article/generative-ai-act-two/

![[assets/slides/9-vGxMoUM9Y/slide-005.jpg]]

OCR text:

> wy
> 
> ba ed

![[assets/slides/9-vGxMoUM9Y/slide-006.jpg]]

OCR text:

> Software APIs are deterministic...
> a,
> Sap > eee * r
> ers “ae |
> fae ad ark e™
> fe Nt By - Bm cor
> oe oe sont . 2
> io —
> +: Guerdraite Al

![[assets/slides/9-vGxMoUM9Y/slide-007.jpg]]

OCR text:

> ...but ML Model APIs are not

![[assets/slides/9-vGxMoUM9Y/slide-008.jpg]]

OCR text:

> Use of LLMs is limited
> when “correctness” is critical.

![[assets/slides/9-vGxMoUM9Y/slide-009.jpg]]

OCR text:

> |
> Alex Graveley @ sxe
> @alexgraveley
> Simple LLM technique that helps a lot (but you might not be using): add |
> - aconstraint checker to ensure valid generation. On violation, inject what
> was generated and the rule violation, and regenerate.
> :+ Querdraie Al

![[assets/slides/9-vGxMoUM9Y/slide-010.jpg]]

OCR text:

> Guardrails AI acts as a safety firewall around your LLMs
> Application Logic
> Standard
> LLM Logic
> Prompt
> LLM API
> Raw Output
> Application Logic
> Guardrails AI
> LLM Logic
> Prompt
> LLM API
> Raw Output
> Reconstruct prompt...
> Fail Validation
> Verification Logic
> Pass Validation

![[assets/slides/9-vGxMoUM9Y/slide-011.jpg]]

OCR text:

> an en anit a el
> ae
> i
> \
> H

![[assets/slides/9-vGxMoUM9Y/slide-012.jpg]]

OCR text:

> 2.
> 
> Guardrails Al under-the-hood
> : Select type of ovtput to waldate : : + UA wT . :
> He career : : ¢ ce er ‘ :
> ae ; LR ot
> 1) Dphawtig Medel | ; t
> tees 5 “t i ‘ [tM Ont in»
> 1 Seg ; ~ etclag eer >! 2 ST — |
> 4B SGeavedi Sranatavenevanenare”™ Z from wer teens Sem i—— Return tpt
> : & ® ' ze ty i 4
> ne CLA Callable : : Inveld vald : :
> te ‘ : ‘ * ’ i % :
> i gc : ao es ;
> 1 wetnetions : Lo emi Be PR Re nee :
> 
> -. Guardraiie Al
> 
> | ;

![[assets/slides/9-vGxMoUM9Y/slide-013.jpg]]

OCR text:

> What Guardrails Al does
> 
> Guardrails Al is a fully open source library that offers
> Framework for creating custom validators
> 
> Orchestration of prompting > verification > re-prompting
> 
> | Library of commonly used validators for multiple use cases
> Specification language for communicating requirements to LLM
> +> Guerdraiie Al

![[assets/slides/9-vGxMoUM9Y/slide-014.jpg]]

OCR text:

> What Guardrails Al does
> | nY|s Guardrails Al is a fully open source library that offers
> Framework for creating custom validators
> Orchestration of prompting > verification + re-prompting
> Library of commonly used validators for multiple use cases
> Specification language for communicating requirements to LLM
> Quartets Ai
> , Se) Seesjo
> i bs

![[assets/slides/9-vGxMoUM9Y/slide-015.jpg]]

OCR text:

> Whynotusepromptengineeringorabettermodel?
> AIE
> Controlling with prompts (including
> LLMs are stochastic:Same inputsSame outputs
> Prompts don't offer guarantees: LLMs don' always follow instnuctions
> AUTOgPt
> Smol°

![[assets/slides/9-vGxMoUM9Y/slide-016.jpg]]

OCR text:

> SUMMIT

![[assets/slides/9-vGxMoUM9Y/slide-017.jpg]]

OCR text:

> i OS
> Implementing Guardrails
> . . Traditional
> Grounding via LLM self
> external systems ML methods reflection
> Rules-based High precision
> heuristics DL classifiers
> :- Gasurdvaite Al

![[assets/slides/9-vGxMoUM9Y/slide-018.jpg]]

OCR text:

> ti

![[assets/slides/9-vGxMoUM9Y/slide-019.jpg]]

OCR text:

> Example: Internal chatbot with “correct” responses
> Problem
> Build a chatbot over the help center
> articles of your mobile application
> © Guntais Ns

![[assets/slides/9-vGxMoUM9Y/slide-020.jpg]]

OCR text:

> How do! prevent LLM hallucinations?
> foes
> ; 5 - COTES
> ing: \ Provenance Guardrails
> y OZ __
> 5 = ;
> in a 2 | Every LLM utterance should have a
> _— source of truth.
> ‘ay (UU ese
> | ‘ a nips. docs guardraisa comapi reference vahdators'squardrails vahdators Provenance 1

![[assets/slides/9-vGxMoUM9Y/slide-021.jpg]]

OCR text:

> e e ee ”
> Example: Validating “correctness
> Appicaton Loge
> Guasdraits Al
> y
> UML How do | change my password?
> - | penne eet
> an Fal Validetion Raw LLM Output
> Verihcanon'viog 2. Log into your account.
> Lore 2.Go to user settings by |
> clicking on the top left
> SI - 6% ie
> 3. Click change password.
> T - °
> Pass Validation
> | -. Querdraie Al |
> l ;

![[assets/slides/9-vGxMoUM9Y/slide-022.jpg]]

OCR text:

> More examples of validations
> e Make sure my code is executable
> 
> e Never give financial or healthcare advice
> 
> e Don’t ask private questions |
> e Don’t mention competitors
> e Ensure each sentence is from a verified source and is accurate
> 
> e No profanity is mentioned in text
> 
> e Prompt injection protection
> 
> e Never expose prompt or sources
> 
> > Querdraiie Al

![[assets/slides/9-vGxMoUM9Y/slide-023.jpg]]

OCR text:

> Learn more
> 
> @ Github:
> 
> @ Website:
> 
> @ Twitter: @ShreyaR or @guardrails_ai
> [a] C pou [a]
> ae is ms Eh
> ag bettors
> aes
> 
> | :. Querdraite Al (alt : an ae.

## Slide-Derived Subjects To Review
Subject extraction uses video title, related session titles/descriptions, transcript context, and OCR text when available. OCR is best-effort and should be reviewed against the embedded slide images.