💡 Inspiration
As AI systems increasingly power global products — from cross-border hiring platforms to international collaboration tools — we realized something critical was missing:
Teams have almost no visibility into how AI behaves differently across regions.
The same AI model can:
- cost significantly more in one country than another,
- hallucinate or degrade in quality for certain regions,
- silently violate compliance or fairness expectations,
- fail without any clear audit trail.
Inspired by VisaVerse’s vision of borderless opportunity and global mobility, we built ObservAI to act as a trust layer for global AI systems — making AI decisions transparent, auditable, and fair no matter where users are located.
🚀 What ObservAI Does
ObservAI is an AI observability and governance platform that monitors every AI interaction across regions in real time.
It helps teams:
- 📊 track AI performance (latency, failures)
- 💰 understand AI cost per region
- 🧠 detect hallucination, toxicity, and quality risks
- 🌍 audit AI behavior across borders
- 🚨 receive real-time alerts when AI systems misbehave
All of this works as a drop-in SDK for Gemini / Vertex AI, with Datadog-powered observability under the hood.
🛠️ How We Built It
ObservAI was built using a production-grade observability stack designed for modern AI systems.
Architecture Overview
$$ \text{User Application} \;\rightarrow\; \text{ObservAI SDK powered by Gemini 3 Flash on Vertex AI} \;\rightarrow\; \text{Antigravity OpenTelemetry Layer} \;\rightarrow\; \text{Datadog Observability Platform} \;\rightarrow\; \text{ObservAI Insights & Governance Dashboard} $$
Core Technologies
- AI Model: Gemini 3 Flash
- Instrumentation: Antigravity + OpenTelemetry (OTLP)
- Observability Platform: Datadog
- Backend: Supabase (Edge Functions + PostgreSQL)
- Frontend & Hosting: Vercel
- Language: TypeScript
Each AI request is automatically enriched with metadata like:
- region
- latency
- token usage
- cost
- hallucination risk
- safety signals
These signals are streamed to Datadog, where we define custom dashboards, detection rules, and alerts.
🚧 Challenges We Faced
- AI-specific observability is fundamentally different from traditional infrastructure monitoring
- Capturing hallucination and safety signals required custom logic beyond standard metrics
- Making observability work reliably in serverless production environments
- Designing alerts that are actionable instead of noisy
- Mapping AI behavior to real-world cross-border risks
We overcame these by treating AI as a first-class risk system, not just another API call.
📚 What We Learned
- Traditional APM tools alone are not enough for AI systems
- AI observability must include quality, safety, and fairness signals
- Developers want tools that are invisible, automatic, and low-friction
- When combined with Datadog, AI telemetry becomes deeply actionable
- Global AI products need auditability by default, not as an afterthought
🌱 Impact & Future Vision
ObservAI enables teams to:
- build safer global AI products
- ensure fairness across regions
- reduce AI costs intelligently
- comply with emerging AI regulations
- trust AI systems at scale
Our long-term vision is to make ObservAI the trust infrastructure for global AI, helping people work, move, and collaborate across borders — safely and transparently.
Built With
- datadog
- gemini
- google-cloud
- npm
- opentelemetry
- react
- supabase
- tailwind
- vertex-ai
- vite
Log in or sign up for Devpost to join the conversation.