We're excited to announce our newest challenge with Sanity!
The Sanity Challenge runs September 18 to October 4. Build an AI agent on structured c...
For further actions, you may consider blocking this person and/or reporting abuse
Just when I thought I was out, they pull me back in.
😂
😅same
That's what she said!
I think you are the best
Let's goooo.
I'm going to participate in this, no matter what!!
I'm so excited!!!
Entered both paths, on one shared dataset: an Oracle → Azure migration knowledge graph with 446 documents, where every claim points at a source.
Path One: Keyword search found the facts but lost the sources. An agent over two Sanity Context endpoints, with a blind four-way eval. Keyword search got 14 of 18 verdicts right but cited the required publishers only 4 times out of 16. The GROQ agent cited them 12 times and grounded all 18 answers.
Path Two: The Legacy Obituaries. Every Oracle feature that didn't survive the move to Azure gets a newspaper obituary: born in, cause of departure, survived by. An App SDK desk certifies the notices, and a Workflows lifecycle runs over the queue.
Both posts include the agent sessions and what went wrong along the way. Thanks for running this one, and good luck everyone.
I vibe-coded something strange: GROQ & Dagger 🕵️♂️⚡
It's an atmospheric cyber-noir interactive crime board built with Next.js, Sanity, and GROQ. Instead of a static table, it models linked suspect dossiers, physical evidence locker items, testimony contradictions, and sequential timelines.
My favorite part was using GROQ graph dereferencing (
->) so you can select any suspect and instantly isolate every piece of connected clues across the entire dashboard in a single query!I NAVIGATED some tricky Vercel deployment hurdles along the way. Check out the full breakdown and live demo here:
👇👇👇
dev.to/inkwavebytes/groq-dagger-un...
Path one submission: TrueStay — a resort-fee desk that adds the fee search hides.
Official pages say “no hidden fees,” “$55,” and “waived for Gold” at once. Not Gold + self-park → $55 + $20. Gold+ booked direct → $0 resort fee, parking still $20.
Writeup: dev.to/ravi_saxena_6fd4d0ef1c9cb/t...
Demo (no login): truestay-five.vercel.app
Entered both paths.
Path One — Errata Desk: dev.to/luisprimecore/errata-desk-s...
Live: luiscore.com/errata-desk/ · check: luiscore.com/errata-desk/check
Path Two — The Needle: dev.to/luisprimecore/the-needle-on...
Dial: luiscore.com/errata-desk/needle
Battery 10/10. decidedBy stays empty until a person signs.
Super excited for this challenge! Sanity + AI agents is such a strong combo. Can't wait to see what everyone builds! 🚀
LEARN. BUILD. REPEAT 🔁
thinking about it.
Yee Awesome lets goo
Boom!! let's craft it!
On it! Standby.
Lets try this. I have no idea how it will be but I'd love to grow and experience.
Interested ! I'm innnn
Let`s Go Baby !!
This is very very very good project 😍😍
locked in.....🎯
That "rough app with honest writeup beats polished one" bit is refreshing. Been there, spent way too long on polish and barely explained it.
Hey ho, let's go!
Wicked to be here thank you mucho 🥰
Okay
:]
Yeppp !! Sanity is amezing.
Hello
Hi to All Devs, Let's Roll...
Gonna give this a shot good luck everyone!
Hello I'm interested
yep thats solid
I'm going to participate in this, no matter what!!
I'm so excited!!
Je vais participer
Entered both paths — demos need no login.
Path One — Errata Desk: dev.to/luisprimecore/errata-desk-s...
Live desk: luiscore.com/errata-desk/ · checks: luiscore.com/errata-desk/score
Project
gsu7qzk9, public dataset.Path Two — The Needle: dev.to/luisprimecore/the-needle-on...
Dial: luiscore.com/errata-desk/needle · workflow: luiscore.com/errata-desk/workflow
Curious what people think of the June 3 split — paper and Magic Online have switched, Arena has not. Same sentence, three answers.
Path Two Submission: Contradiction Triage
I built a contradiction-triage workflow using the Sanity App SDK.
The idea is pretty simple: when two company documents disagree, an agent reviews the conflict and proposes a ruling with supporting precedent. A human can then approve or override that recommendation. Once a decision is made, it becomes a typed, queryable document that future decisions can reference.
One important part of the design is that the
ruledtransition can only be triggered by a human. That rule is enforced structurally in the app rather than relying on convention.Links
cqb58l0fsanity.judges@tokmail.net/Shahbaz123What I Learned
The most interesting part of the project ended up being the bugs I found along the way.
Three bugs initially looked unrelated, but they all came back to the same underlying issue: something had one meaning in one place and a different meaning somewhere else.
All three bugs made it past
sanity build.I only caught them after running
tscand testing the actual GROQ query against the live data.That ended up being one of my biggest takeaways from the project:
The type checker and live queries caught issues that the build alone didn't.
Happy to answer any questions or walk through the implementation.
Live Demo: sanity-challenge.vercel.app
Collection Demo (GROQ live): sanity-challenge.vercel.app/collec...
Sanity Studio (public, no login): sanity-challenge.vercel.app/studio
GitHub (MIT): github.com/dnysaz/niche-collector
Sanity Project ID: 71m89sy5
Dataset: production — 71m89sy5.apicdn.sanity.io/v2024-01...
Organization ID: os9xuj7u1
Submission tag: #sanitychallenge
Path Two submission: AI Director — High-Performance Commercial & SaaS Engine Powered by Sanity CMS
Writeup: dev.to/aisiteflow/ai-director-high...
Live Demo: ai-director-roan.vercel.app
Commercials Showcase: ai-director-roan.vercel.app/commer...
SaaS Templates: ai-director-roan.vercel.app/templates
GitHub: github.com/aisiteflow/ai-director
Sanity Project ID: ix5izt37 (Dataset: production)
The 150 KB document limit makes this almost impossible for any practical workflows. I spent a lot of time in on this only to get an error.
(I am been in the contact with the great Sanity team on their discord.) But its just feels like a waste of time spent. :(
Sanity community via discord*
The part I'd copy is where the authority sits. The model gathers evidence and explains, and a few lines of plain code make the call. I build coding-agent workflows the same way: the agent proposes, and a fixed gate decides what gets merged.
Your mid-run recovery made me think about the rule's inputs. When a tool drops out halfway, client_approved or director_flagged can end up unknown for some shots. I'd have the filter treat unknown as ineligible and have the note say which fields were missing. Then a partial run still gives a safe answer, and the person on call knows how much the system actually saw.
Did you think about having the scoring function also return why each shot was excluded? That list seems useful for debugging and for the dashboard note.
Path one submission: Nebula Agent - reads the errata the rulebook hides.
Rulebook says 5 energy tokens, 10 stars. Errata says 8 and 12. Strategy guide cites the errata but still mentions 10. All three live in the same dataset. A keyword search returns all of them with no way to rank.
My agent queries Sanity, sees the contradiction, and replies: "Rulebook says 5. Errata corrects it to 8. Here's the source URL for each."
The magic isn't the LLM, it's that every document has a structured 'source' field, so the agent can attribute each claim to its origin.
Writeup: [dev.to/vicarioy/nebula-agent-an-ai...]
Code: [github.com/Vicarioy/sanity-nebula-...]
green ticks close the UX. settlement week still needs a tip anyone can GET after the portal flips.
which public query proves the cut without a second shadow ledger?
curious what breaks first: MDR opacity or hop finality.
marker1001h1828-dt
😀
I am the judge/mentor here.
goona hit first
dev.to/kaushik_patil_b20f74fe212/b...
i thought we were having a break from the bloat? i'm just vibin with registers and opcodes
Hii
Hi
Hi there!!! Wanted start my career in IT backend, any advice will be okay!!
我是中文文字组合内容，和人工智能产品，对话方式的，使用者，及对工具的属性。非工具性属性，有好奇心，的追述者
hello
Hi 😌