Juan Peredo

Juan Peredo is a software architect, AI entrepreneur, and creator of Bolbeck Imagine, a platform for producing illustrated, animated, multilingual children’s stories. His work makes generative AI accessible to families, educators, and authors while confronting the operational challenges of building dependable applications.

Peredo began his career developing software to optimize truck routing in open-pit mines. He subsequently worked across cloud architecture, consulting, and software delivery, with experience at Amazon Web Services, Strategy&/PwC, and Booz & Company. His open-source Go and Svelte microservices application, built for Docker and Kubernetes, reflects his background in distributed application architecture.

Through Bolbeck Imagine, Peredo combines storytelling with generated illustrations, animation, narration, translation, and interactive experiences. His approach applies production engineering to creative tools intended for people without specialized AI expertise.

• Production AI requires operational discipline. Model selection, hallucination mitigation, infrastructure, evaluation, and monitoring remain necessary after an application’s initial code has been generated. Models must be tested against their actual use case as their behavior changes.

• Retrieval quality depends on information quality. Retrieval-augmented generation cannot compensate for outdated or inaccurate source material; guardrails, output validation, and fine-tuning introduce their own costs and limitations.

• Externalized prompts enable expert collaboration. Managing prompts outside application code lets educators and other specialists improve instructions directly while helping teams adapt to changing models.

• Agent economics shape product viability. As Peredo illustrates in his AI Engineer Summit presentation, model choice, repeated tool calls, and parallel execution determine an application’s cost and responsiveness.

• Agent observability makes failures diagnosable. Tracing model calls, inputs, outputs, and errors exposed a case-sensitive user lookup that Peredo corrected by updating an externally managed prompt.