Giran Moodley is a founding field engineer at Braintrust, building the AI evaluation and observability company’s presence across Europe, the Middle East, and Africa. He helps organizations turn promising language-model prototypes into production systems whose behavior can be traced, evaluated, and improved.
Moodley studied at the University of Birmingham and worked as a senior solutions consultant at PagerDuty, advising organizations on service ownership, incident response, and operational architecture. He also built PagerDuty Live, an open-source interface for managing incidents in real time, alongside projects involving Terraform and AWS Lambda integrations.
At Databricks, he expanded his focus to enterprise data platforms and generative AI. His 2024 guide to fine-tuning language models, coauthored with Ellen Hirt and Narjes Majdoub, examines data preparation, training, evaluation, deployment, and monitoring. It also cautions that fine-tuning is not always necessary: prompting, retrieval-augmented generation, smaller models, or existing models may better fit an application’s data, latency, and cost requirements.
After working at Databricks through 2025, Moodley joined Braintrust to establish its field-engineering presence across EMEA. His work concentrates on several practical challenges:
- Production observability beyond logs: Trace nested model calls, tool invocations, inputs, outputs, latency, token usage, and failures together to understand how an AI application reached a decision.
- Multi-stage agent workflows: Break complex behavior into inspectable steps such as collecting context, triaging requests, checking policy, drafting responses, and escalating sensitive cases to humans.
- The evaluation flywheel: Turn production failures into regression cases; combine representative datasets, inexpensive deterministic checks, and model-based judgments; then measure whether changes improve behavior before deployment.
- Cost-aware operational governance: Use versioned prompts, managed configuration, reproducible experiments, and selective evaluation sampling to balance collaboration, traceability, and inference costs.
At AI Engineer Europe 2026, Moodley demonstrated these practices in a joint Braintrust–Trainline workshop. Trainline engineers described their own travel-assistant systems; Moodley focused on the instrumentation, evaluation, and production feedback mechanisms needed to operate complex AI applications reliably.
