Experience
Entel Perú
RAG automation for 100K+ documents per month, 84s-to-1s audio inference, LLM evaluation, and AWS serverless AI infrastructure.
Built end-to-end GenAI systems at Entel spanning RAG, high-throughput inference, model evaluation, and serverless AI infrastructure. The merged role combines the official LinkedIn title progression with the resume-level outcome story: 100K+ automated documents per month, 80% less manual processing, about 84 seconds of audio processed per second of compute, and a 25% increase in internal chatbot satisfaction.
Designed an end-to-end RAG pipeline on AWS Bedrock using LangChain and vector search to automate classification for 100K+ documents per month and cut manual processing by 80%.
Built a privacy-preserving Whisper transcription pipeline that processed about 84 seconds of audio per second of compute using GPU-optimised ECS, batch processing, autoscaling, Step Functions, Glue/Spark, and S3.
Selected assessor summaries over full audio transcripts after measuring that audio cost about 4.5x more input tokens for only about 4 percentage points of precision gain.
Developed an LLM evaluation framework across relevancy, correctness, latency, and cost, with monitoring and feedback loops that increased internal chatbot satisfaction by 25%.
Led serverless AWS architectures for GenAI microservices and data pipelines, adding ECS, RDS, S3, CloudFormation, CloudWatch, Bitbucket Pipelines, Spark, Azure Active Directory, and FastAPI where needed for platform scale and access control.
100K+
Documents per month
80%
Less manual processing
~84s/s
Audio per compute second
+25%
Chatbot satisfaction