Caeliq
Prototype: AI agent proves fully automated travel proposal creation - request to PDF.
The Challenge
An observed bottleneck in business travel consulting: creating a single travel proposal required 15 to 60 minutes of manual work - searching, comparing, formatting, assembling. With a high volume of requests, the effort scaled linearly, with no structural lever for efficiency gains.
Data sovereignty was a non-negotiable requirement: client data could not leave the agency's own infrastructure. The solution had to run on-premise or at minimum give full control over the data path - cloud systems were ruled out for compliance reasons.
Our Solution
The prototype proves the automation logic end-to-end: an AI agent receives a travel request, searches a structured flight database, selects suitable options, and outputs a formatted PDF proposal - without manual intervention. The database is currently simulated, not yet connected to a live GDS; it serves to prove that the agent logic works correctly with real structured data.
The agent is built on an open-source coding agent fundamentally reconfigured for this workflow: different domain, different toolset, robust sandbox architecture. It can only execute predefined scripts - no arbitrary code execution. Its behavior is strictly bounded to its assigned scope and fully controllable.
The application is designed for multi-tenant operation and runs containerized via Docker - locally on a single device or on a shared server. The AI backend is provider-agnostic: cloud LLMs or local models, depending on data privacy requirements. The next defined step is GDS integration through a pilot partnership with a business travel agency.
Outcomes
System Architecture
Request intake & task queue
Parallel, isolated agent session per request
Flight search & proposal selection from structured database
PDF assembly & output
Edit loop: agent-assisted revision on demand
Technology Stack
Backend
Frontend
KI-Agent
Inference
Infrastruktur
01
Local AI Server
Self-contained AI infrastructure: Run capable local AI models without external API dependencies.
A system built on prosumer hardware capable of running high-capability AI models up to 120 billion parameters at high speed - two inference backends, one server, entirely without the cloud, no external costs, no limits.
Ready to take the first step?
30 minutes. No pitch. You'll leave with clarity - not another proposal.
Free discovery call · No commitment