Jenny Truong

• Audited multimodal annotations for Meta's SAM 2 foundation model catching labeling errors in image and video data before they degraded ground truth quality. • Caught systemic annotation errors upstream identifying vendor-level labeling trends early and escalating before they compounded in the training dataset. • Bridged vendor teams and internal leadership across 6–10 stakeholders to maintain annotation consistency, align on rubric interpretation, and surface quality flags before they reached the model.

• Designed training data pipelines for a multimodal browser-navigation AI agent scoping UI interaction workflows, building test environments, and authoring SOPs for platforms including Shopify. • Served as sole POC for a 25-person annotation vendor team owning data collection, QA, engineer escalation, and final training package submission. • Dogfooded every workflow before vendor deployment stress-testing agent behavior and refining prompts to maximize training signal quality and capture real UI edge cases. • Delivered an agentic workflow for an enterprise fleet tracking client enabling the AI agent to autonomously resolve container queries that previously required manual cross-platform searches.

• Improved NLP chatbot accuracy across 1,000+ intent categories using confidence score analysis to identify low-performing intents and cleaning mislabeled training data at the source. • Prevented model degradation by synthetically generating training utterances maintaining sufficient signal for each intent class when cleaned datasets fell below viable thresholds. • Owned the full NLP intent optimization cycle from log extraction and dataset cleaning to BigQuery ingestion and production accuracy monitoring.

• Labeled safety-critical computer vision data for Tesla's autonomous driving system maintaining annotation precision across complex real-world scenarios where accuracy directly influenced perception and vehicle safety. • Reduced label noise by surfacing edge cases and rubric gaps enabling engineers to refine annotation guidelines before errors propagated across the dataset.

• Coordinated vehicle logistics and managed documentation workflows • Streamlined handoffs between customers, sales, and operations teams • Maintained process accuracy under time-sensitive conditions