DeSpiral

The AI Circuit Breaker for Health Anxiety

Comment

Inspiration

Excessive and repeated online searching for health and medical information ("cyberchondria"/"health anxiety") are a modern epidemic.

When someone experiences a sudden bodily symptom, their instinct is to search Google or AI (Gemini/ChatGPT). Instead of finding reassurance, they are met with worst-case medical scenarios, triggering an intense panic loop and physical tension.

Reddit Post - Doctor said not to worry but i kept digging in google

Reddit Post - Is there a Chrome addon or other program that will stop me from Googling "that stuff?

Global prevalence estimates for cyberchondria generally range from 30.7% to 55.6% across various adult, student, and patient populations, according to recent PubMed literature.

Traditional search engines fail "health anxious" users because they treat every query as a cold information retrieval task rather than a vulnerable emotional state.

DeSpiral was born to act as a digital "circuit breaker"—an intelligent browser layer that interrupts panic loops, regulates the physical nervous system, and provides calm, empathetic health context before worst-case information is consumed.

What it does

DeSpiral is a Chrome Extension that transforms the web search experience for health-anxious individuals through five core features:

1. 20-Second Somatic Interstitial: When a health query is detected on Google , Gemini or ChatGPT, a full-screen overlay pauses the user with randomized, clinically backed grounding exercises (e.g., 4–6 extended exhale breathing, somatic muscle resets, 3-2-1 visual grounding).

2. Zero-Latency Pre-Fetching: While the user completes their 20-second grounding exercise, DeSpiral streams the AI response in the background. When the timer hits zero, a fully formatted, calm response is waiting.

3. Empathetic AI Triage & De-escalation: Powered by CockroachDB and Claude on AWS Bedrock, DeSpiral provides grounded, non-alarmist medical context, separating emergency red flags from common anxiety-induced symptoms.

CockroachDB powers DeSpiral’s episodic memory and pattern engine by unifying vector search with relational analytics. It stores semantic embeddings of past search episodes to provide cross-turn conversational context during AI triage without losing historical grounding.

AWS serves as the backbone of DeSpiral’s serverless streaming and AI intelligence layer. It utilizes AWS Lambda and AWS Bedrock with Claude 3.5 Sonnet to deliver real-time, empathetic triage and multimodal lab report analysis with guaranteed zero data retention.

4. Stateless Lab Report & Symptom Reconciler: Users can upload lab test reports (PDFs or images). DeSpiral processes the report alongside their recent search history completely in-memory (zero database retention) to show a "Search vs. Reality Check"—proving when normal lab values contradict their panic searches.

5. Patterns & Insights Dashboard: Analyzes search logs using CockroachDB to highlight search triggers, top keywords, and high-risk time windows (e.g., late-night spikes between 11 PM and 1 AM).

How it was built

Frontend Layer: Chrome Extension (Manifest V3) using injected content scripts, CSS animation keyframes for somatic guides, background service workers, and long-lived Chrome Ports.

Backend Architecture: FastAPI running on AWS Lambda utilizing AWS Lambda Web Adapter (LWA) to support real-time HTTP response streaming.

AI & Orchestration Engine: AWS Bedrock invoking Claude 3.5 Sonnet via the converse_stream API for both text triage and multimodal PDF/image analysis.

Database & Memory: CockroachDB for vector storage (RAG), cross-turn episodic memory retrieval, and analytical aggregation queries.

Deployment & Hosting :

For Hackathon Judges - Vercel for the production landing page, user documentation, video embeds, and static .zip extension distribution.

For Target Audience (Future Scope) - Chrome Web Store.

Challenges faced

Safety vs. Utility: Balancing empathetic de-escalation with medical safety required precise prompt engineering, including a mandatory, instant "Medical Emergency / Skip" escape hatch on overlay.

Privacy-First Multimodal Processing: Designing the Lab Report Analyzer required guaranteeing absolute zero data retention. Engineered a completely isolated, ephemeral endpoint that handles base64 PDF payloads strictly in-memory without writing to disk or CockroachDB.

Accomplishments to be proud of

Production-Grade Serverless Stack: Created a full-stack distributed pipeline—combining CockroachDB for scalable vector and relational storage with AWS Bedrock streaming via Lambda Web Adapter (LWA).

100% Agentic Solo Build with Coding Agent: Built end-to-end by a single developer leveraging AI coding agent. From the Manifest V3 Chrome Extension architecture and FastAPI background streams to the complete CockroachDB database schema and vector search setups, the entire codebase was scaffolded, debugged, and shipped through human-in-the-loop AI agent steering. Rather than relying on "vibe coding," development was driven strictly by upfront specifications and PRDs.

Learnings

Unified Relational & Vector Storage in CockroachDB: Discovered the efficiency of pairing structured analytics with unstructured vector embeddings in a single CockroachDB instance, avoiding the overhead of maintaining separate relational and vector databases.

Serverless Response Streaming with Lambda LWA: Learned how to leverage AWS Lambda Web Adapter (LWA) to bypass traditional API Gateway payload and timeout limits, enabling direct, real-time token streaming from AWS Bedrock into Chrome Extension background workers.

Spec-Driven AI Engineering vs. "Vibe Coding": Validated that pairing detailed PRDs and technical blueprints with an AI agent like Claude Code drastically eliminates architectural drift, allowing a solo developer to reliably ship complex distributed systems.

What's next for DeSpiral

iOS / Android Mobile Application : Bring DeSpiral to mobile to intercept panic health searches done from mobile device.

Built With

Share this project:

Updates

Submission history