WKO AI Recommendation Engine
Faster APIs, automation and AI recommendations for WKO Inhouse
For WKO (Austrian Economic Chamber), through its shared service center WKO Inhouse, we built an AI recommendation engine for content, cut API response times from 8 seconds to 0.7 seconds and automated long-running data pipelines.
Published
- Industry
- Government services, Public sector
- Region
- Austria
- Services
- Custom software development, Cloud consulting, AI consulting, API development
- Duration
- 3 years and ongoing
- p99 API response time
- 8 s → 0.7 s
- Ongoing partnership
- 3+ years
- Engineers and data scientists
- 4
Intro
WKO Inhouse GmbH is an independent company of the Austrian Chambers of Commerce and the service provider for all of its departments, covering IT solutions, data center services and more.
They came to us with slow APIs and unreliable data-transfer workflows. Within two weeks we brought API response times from 8 seconds to under 0.9 seconds; the 99th percentile has since settled at 0.7 seconds. We also automated long-running tasks and put the cloud infrastructure on a scalable, maintainable footing — work that has continued for more than three years.
Goals and challenges
- 01
Performance
Users often waited around 10 seconds for API responses, frustrating them.
- 02
Job automation
Long-running data transfer processes and broken scripts needed constant manual intervention.
- 03
Infrastructure
Infrastructure was scattered and deployed inconsistently, which made it hard to maintain.
Solution
Performance improvement
We refactored the FastAPI application and optimized its queries and architecture. Within two weeks response times were under 0.9 seconds; 99th-percentile response times now stand at 0.7 seconds, down from 8 seconds.
Job automation
GCP Cloud Functions and Cloud Run Jobs now schedule and run tasks such as reindexing data, updating databases and publishing messages to queues.
Infrastructure as Code
Terraform gives reliable, repeatable deployments across development, testing and production, with Cloud Build for CI/CD and GCP Monitoring with custom alerts.
Security and compliance
We ran a security audit, patched the vulnerabilities it found and secured data transfer between BigQuery, Firestore and PostgreSQL.
Content recommendations
We also built a content recommendation engine that suggests related articles.
Team
- 2 Software Engineers
- 1 DevOps Engineer
- 1 Data Scientist
Results
- 99th-percentile response times dropped from 8 seconds to 0.7 seconds.
- WKO Inhouse said our speed of delivery and the quality of the work exceeded their expectations.
- Collaboration ran mainly over email and messaging apps; WKO Inhouse highlighted our flexibility and friendly communication as major strengths.
- Project duration
- 1 week
- Data scientists
- 2
Intro
WKO wanted to make its YouTube channel easier to explore with personalized recommendations. We built a Python service that uses OpenAI to analyze video metadata — descriptions, tags and categories — organize the catalogue automatically and enable tailored video discovery.
Goals and challenges
- 01
Reliable interpretation
Prompts had to be precise enough for the model to interpret each video reliably.
- 02
True-to-content recommendations
Recommendations had to match what a video is actually about.
- 03
Robust ranking
The ranking behind personalized suggestions had to stay robust across the catalogue.
Solution
The service cleans descriptions, sharpens category definitions with detailed German context and runs tailored German-language OpenAI prompts with several query variants for robust categorization.
Data preprocessing
- Cleaning descriptions by removing promotional text and links
- Enhancing category definitions for clearer AI context
- German-language prompts aligned with the audience
Techniques
- OpenAI API with tailored German prompts
- Multiple queries combining title, description and tags
Team
- 2 Data Scientists
Results
- Structured content organization and easier navigation of the channel.
- A personalized experience that connects viewers with relevant videos.
- A recommendation-ready pipeline that helps WKO surface relevant content and improve engagement.
The work is completed quickly and with good quality. Their friendly communication and flexible approach are unique. The team is thorough in their security reviews and focuses on delivering tangible improvements that we can measure and see in our system’s performance.
Project details
- Client
- WKO Inhouse GmbH (Austrian Chambers of Commerce)
- Industry
- Government services, Public sector
- Region
- Austria
- Services
- Custom software development, Cloud consulting, AI consulting, API development
- Technologies
- Python
- FastAPI
- GCP (Cloud Functions, Cloud Run, BigQuery, Firestore)
- Terraform
- PostgreSQL
- Cloud Build
- OpenAI
- Team
- 4 — 2 software engineers, 1 DevOps engineer, 1 data scientist
- Duration
- 3 years and ongoing
