How Omio is building the future of conversational travel
Key Points
- Connected OpenAI to 3,000+ transportation providers
- Reduced product development effort to ~20%
- Cut project cycle time from ~3 months to ~1 month
Summary
Omio integrated OpenAI models (ChatGPT, Codex, API) with live transportation inventory and booking systems to create conversational travel experiences and transform internal workflows. Externally this enables grounded, bookable itineraries from natural-language queries; internally Codex is embedded across the SDLC and custom connectors surface systems and data to AI tools. Results include 3,000+ connected providers across 47 countries, a ~80% reduction in product development effort, and cycle times shortening from a quarter to roughly one month.
Key Points
- Architecture: OpenAI models are grounded in live inventory/pricing and booking APIs so responses are verifiable and actionable.
- Internal adoption: ChatGPT for experimentation; Codex embedded across research, coding, testing, reviews, monitoring, and maintenance.
- Integrations: Custom connectors expose internal systems and data to AI tools to move from retrieval to execution.
- Metrics: 3,000+ transportation providers, 47 countries, products built in ~20% of previous effort, projects shortened from ~3 months to ~1 month.
- Governance: Human-in-the-loop model — AI accelerates work while people retain responsibility and accountability.
Technical considerations
- Grounding layer: ensure responses are tied to live data and include provenance for bookings.
- Connectors & auth: build secure connectors, enforce RBAC, and audit model-driven actions.
- Observability: instrument latency, error rates, and booking success; plan fallbacks and human review paths.
- Developer workflow: integrate Codex into CI/CD, code review, and testing pipelines to retain quality while accelerating delivery.
Actionable takeaways for engineers
- Prototype a conversational endpoint that binds model output to a verified pricing/inventory API.
- Build modular connectors and a grounding service to validate and sign-off AI responses before execution.
- Pair broad developer access with guardrails: logging, approvals, and automated tests to maintain accountability.
- Measure impact with dev-effort, cycle-time, and production reliability metrics before broad rollout.