DEV Community: Aastha Yadav

From Prompts to Product: Building StudentOS with Google Antigravity + Sanity

Aastha Yadav — Fri, 02 Oct 2026 09:46:19 +0000

StudentOS is an AI-powered student productivity and opportunity intelligence platform built to help students move from discovering an opportunity to actually becoming ready for it.

For this challenge, I extended StudentOS with an Opportunity Intelligence experience powered by Sanity.

The system combines structured opportunity knowledge — opportunities, organizations, eligibility rules, skills, application processes, resources, and sources — with private student context such as skills, goals, education, and progress.

Students can select an opportunity and see their fit, eligibility, skill gaps, application requirements, and personalized preparation guidance.

The goal is simple: instead of only asking "What opportunities are available?", StudentOS helps answer "Am I ready for this opportunity, and what should I do next?"
Live Demo: YOUR_DEPLOYED_WEBSITE_URL

Demo Video: YOUR_YOUTUBE_VIDEO_URL

The demo showcases the Opportunity Intelligence workflow, including Sanity-powered opportunity knowledge, readiness assessment, skill-gap analysis, application requirements, and the grounded AI Q&A experience.
GitHub Repository: https://github.com/aastha-yadav2/studentos-ai
I built and iterated on StudentOS using Google Antigravity as my AI-native development environment.

Rather than rebuilding the application from scratch, I used Antigravity to extend an existing StudentOS codebase and progressively introduce a Sanity-powered opportunity knowledge layer.

1. Starting from the existing product

StudentOS already had student context, opportunity discovery, matching, applications, and preparation features.

The first goal was to understand the existing architecture and identify where Sanity could add meaningful structured knowledge instead of creating a separate application.

2. Designing the Sanity knowledge model

I prompted Antigravity to design a structured content model around:

Opportunities
Organizations
Skills
Eligibility Rules
Application Processes
Resources
Sources

The important design decision was to connect these records rather than store everything as one large opportunity document.

This created relationships such as:

Opportunity → Organization → Eligibility → Skills → Application Process → Resources → Source

3. Building the Opportunity Intelligence layer

I then prompted Antigravity to connect the Sanity knowledge layer with the existing StudentOS opportunity system.

The implementation combines:

Sanity opportunity knowledge
Student profile/context
deterministic eligibility
existing 50/30/20 matching
AI-powered reasoning and preparation

This resulted in an Opportunity Intelligence drawer where students can understand their readiness for a specific opportunity.

4. Where the build got stuck

The first major issue appeared when an opportunity that was not present in the live Sanity dataset fell back to an existing GSoC record.

This caused unrelated GSoC information to appear for another opportunity.

Instead of hiding the problem, I traced the data flow and used Antigravity to implement strict opportunity identity validation and opportunity-specific fallbacks.

The final system rejects mismatched Sanity documents instead of silently displaying another opportunity's content.

5. Debugging the AI Q&A

A second issue appeared in the Opportunity Intelligence Q&A.

Different questions such as:

"Am I actually ready?"
"What documents do I need?"
"What skills am I missing?"
"How can I prepare over the next 2 weeks?"

were initially returning the same generic response.

I traced this to the AI Router request validation. The opportunity_qa request type had not been registered in the Supabase Edge Function, so requests were failing before reaching the intended AI flow and falling into a generic fallback.

I corrected the request routing and then added:

deterministic question-intent classification
opportunity-specific AI context
response validation
anti-cross-opportunity leakage checks
deterministic intent-specific fallbacks
automated Q&A tests

6. Iterative validation

The final implementation was tested by switching between different opportunities and checking that their eligibility, skills, application information, resources, and Q&A context remained isolated.

I also verified the production build and Sanity Studio build after the changes.

The result was not just a generated UI. The build evolved through several rounds of prompting, inspection, debugging, testing, and course correction.
Sanity Project ID: p2hu7iqp

Dataset: production

Sanity is used as the structured content layer for the Opportunity Intelligence experience, with connected records for opportunities, organizations, eligibility rules, skills, application processes, resources, and source provenance.

Agent Session

The project was developed using Google Antigravity as the AI-native development environment.

A DEV Agent Session transcript is not included because Google Antigravity is not currently listed among the supported Agent Sessions upload sources.

StudentOS Opportunity Intelligence: A Sanity-Powered AI Agent for Opportunity Readiness

Aastha Yadav — Fri, 02 Oct 2026 09:30:53 +0000

This is a submission for the Sanity Challenge, Path One: Ship an Agent That Queries Real Content

What I Built

StudentOS Opportunity Intelligence is an AI-powered opportunity readiness agent built into StudentOS.

Instead of simply showing students a list of internships, hackathons, fellowships, and developer programs, StudentOS answers a more useful question:

“Am I actually ready for this opportunity?”

The system combines two types of context:

Private student context — skills, education, goals, applications, and progress.
Structured Sanity knowledge — opportunities, organizations, eligibility rules, required skills, application processes, documents, resources, and source provenance.

StudentOS then evaluates the student's deterministic eligibility and skill fit, identifies gaps, and uses grounded AI to answer opportunity-specific questions.

For example, a student can select an opportunity and ask:

Am I actually ready for this?
What documents do I need?
What skills am I missing?
How can I prepare over the next 2 weeks?

The result is an actionable path from opportunity discovery → eligibility → skill gaps → preparation → application.

Demo

Live Project: https://studentos-ai-phi.vercel.app/app

Demo Video: https://youtu.be/jMnrf-GSKoA

The demo focuses on the Opportunity Intelligence flow:

Select an opportunity.
Retrieve its structured knowledge from Sanity.
Calculate deterministic student-opportunity fit.
Show eligibility and skill gaps.
Surface the application workflow, documents, and resources.
Ask the grounded Opportunity Intelligence Agent follow-up questions.
Switch between opportunities and verify that the agent uses the newly selected opportunity's context.

Code

GitHub Repository: https://github.com/aastha-yadav2/studentos-ai

The repository contains the StudentOS application, Opportunity Intelligence implementation, Sanity integration, AI Router integration, fallback handling, and validation logic.

How I Used Sanity

Sanity acts as the structured knowledge layer behind StudentOS Opportunity Intelligence.

I modeled opportunity knowledge as connected structured content rather than keeping application information in a single unstructured prompt.

The Sanity dataset contains structured records for:

Opportunities
Organizations
Skills
Eligibility Rules
Application Processes
Resources
Sources / provenance

An opportunity is connected to its organization, eligibility rules, required skills, application process, resources, and source information.

StudentOS retrieves the selected opportunity's knowledge graph and combines it with the student's private context.

The deterministic matching layer calculates the student's fit using the existing 50/30/20 scoring model, while deterministic eligibility remains authoritative.

The AI layer is then grounded in the selected opportunity's resolved Sanity context.

This distinction is important: Sanity provides the opportunity knowledge; StudentOS provides the student context; the agent combines both to produce actionable readiness guidance.

I also added defensive identity validation and opportunity-specific fallbacks so that information from one opportunity cannot silently appear while another opportunity is selected.

For example, switching from Hack2Skill to Google Summer of Code changes the complete opportunity context used by the agent, including eligibility, skills, documents, application workflow, resources, and Q&A.

The Q&A layer also classifies questions into intents such as:

Readiness
Documents
Skills
Two-week preparation
General opportunity questions

This allows the agent to provide different, grounded answers rather than returning the same generic response for every question.

Sanity Project Details

Sanity Project ID: p2hu7iqp

Dataset: production

The live Sanity project contains the structured opportunity knowledge used by StudentOS Opportunity Intelligence.

The project currently contains the opportunity knowledge graph and its related organizations, eligibility rules, skills, application processes, resources, and source records.

Agent Session

The project was developed iteratively using Google Antigravity as the AI-native development environment.

Antigravity was used throughout the build to implement, debug, test, and refine the StudentOS Opportunity Intelligence experience, including the Sanity integration, structured opportunity knowledge graph, deterministic matching, grounded Q&A, opportunity isolation, and fallback handling.

An Agent Session transcript is not included because the development session was conducted in Antigravity rather than one of the session sources currently supported by the DEV Agent Sessions uploader.

The final implementation specifically includes:

Sanity-backed opportunity retrieval
Structured opportunity relationships
Deterministic eligibility and matching
Grounded AI reasoning
Opportunity identity validation
Opportunity-specific fallback handling
Q&A intent classification
Anti-cross-opportunity response validation
Opportunity switching isolation
Automated Q&A tests

Why Structured Content Matters

The core of this project is not simply asking an LLM about opportunities.

The opportunity information is modeled as structured, connected knowledge.

That allows StudentOS to reason over specific relationships such as:

Opportunity → Organization → Eligibility → Skills → Application Process → Documents → Resources → Source

The agent can therefore answer questions using the selected opportunity's actual structured context instead of relying on a generic prompt or a flat keyword search.

StudentOS turns that structured knowledge into something directly useful to a student:

Discover the opportunity. Understand your fit. Identify the gaps. Prepare. Apply.