THIN

Publication Date:

Learn how we helped Thincrs, the professional learning startup, migrate their OpenAI operation to Amazon Bedrock reducing errors by up to 80%.

THE CHALLENGE

By the time it approached ITERA, Thincrs was facing several technical challenges that compromised its value promise.
Content generation with GPT-4o had high latency and inconsistencies, which affected both quality and speed of delivery; In addition, the existing architecture did not allow for effective personalization of content per customer, which limited the relevance of the solutions offered. Likewise, the lack of data segmentation also generated concerns about privacy and information security; and finally, the operation based on asynchronous flows with multiple external dependencies was difficult to scale and complex to maintain, which put customer retention at risk due to the low quality of service.

WHAT WE DID

We implemented a new architecture based on Amazon Bedrock, combining advanced artificial intelligence tools with secure, scalable, and highly available infrastructure. The main services used were:

  • Amazon Bedrock: Custom agents with RAG (Retrieval Augmented Generation).
  • Amazon OpenSearch: Customer-segmented vector base.
  • Amazon Lex: Conversational chatbot integrated with Bedrock.
  • AWS Lambda and API Gateway: Serverless backend ready to scale.
  • Amazon S3: Reliable storage for generated content.
  • AWS IAM, CloudWatch, and CloudTrail: Advanced security, monitoring, and auditing.

In addition, key aspects such as:

  • Observability with log and metric analytics in CloudWatch.
  • Security and privacy through data segmentation, encryption, and access control.
  • Version control to ensure the quality of prompts and responses.
  • Administrative management from a friendly interface for the validation of results.
RESULTS

The technology transformation achieved a tangible and measurable impact on Thincrs’ operation:

  • 80% fewer errors in the generated content (hallucinations reduction from 50% to 10%).

  • 75% less development time per release (from 1 hour to just 15 minutes).

  • 60% more monthly content, going from 2,000 to 3,200 items.

  • 86% more agility to attend to business opportunities (from 7 months to 1 month).

  • 4% savings per million tokens processed when switching from GPT-4o to Amazon Bedrock.

  • Operational stability with constant latencies (~30 seconds) and elimination of critical failures.

With this transformation, Thincrs consolidated a modern, scalable and highly efficient infrastructure, which allows it to provide a more personalized experience to its customers and sustain its growth as a technology provider in the education sector.

Would you like to revolutionize your processes with generative AI? Write to us and transform your operation with your feet in the cloud.

Ready to take full control of your cloud investment?

Explore the services we have for you.