Robert Biggar

Built a Creative Agent Evaluation Platform for drafting campaign briefs and copy across 40+ account teams. Designed eval harnesses using LLM-as-judge plus human raters with brand, factuality, and compliance rubrics, improving rubric agreement to 85% and catching regressions earlier. Embedded with strategy/client service teams to map briefing workflows and reduce manual research time by 40%. Improved RAG retrieval quality using chunking and reranking with pgvector, increasing grounding scores by 25%. Implemented adversarial test suites for hallucinations and regulated-category risks, blocking failing evals in CI/CD. Developed a React/TypeScript review console for side-by-side creative ratings feeding back into datasets, and scheduled nightly eval runs via Airflow with dashboards for stakeholders. Packaged services in Docker for reliable deployment and debugging.

Architected a RAG assistant over contact center transcripts and CX research, reducing analyst insight turnaround by 50%. Led an evaluation pipeline covering retrieval precision, answer faithfulness, and citation accuracy, using labeled datasets and pytest to raise faithfulness by 20% over three releases. Owned production reliability on AWS by deploying Dockerized workloads on ECS with CloudWatch alarms to sustain 99.5%+ availability. Built transcript ingestion and embedding pipelines with AWS Lambda, PostgreSQL, and pgvector, handling retries and duplicate records. Partnered with CX consultants to scope journey analytics use cases and translate them into prioritized agent features with success metrics. Added PII redaction and access controls before embedding to meet privacy requirements. Designed microservice boundaries using Node.js gateways and Redis caching to keep latency predictable as data grew, and ran A/B tests on prompts and retrieval settings to decide production configuratio

Wrote and tested scripted conversation flows for branded chatbots on Facebook Messenger using JavaScript and Node.js webhooks for campaign launches. Implemented keyword and intent matching with Python and NLTK to improve routing accuracy and reduce fallbacks. Logged conversation transcripts in MySQL and iterated on phrasing/response templates to keep replies on brand. Supported A/B testing of copy and button prompts with the analytics team to inform launch creative decisions. Debugged webhook timeouts and formatting issues with regression checks to prevent broken flows from shipping.