Zahra Shergadwala

About

Built and owned an end-to-end AV operations automation and dashboarding platform for USC event operations, used daily to coordinate hundreds of live events. Delivered a React/TypeScript full-stack system integrating Smartsheet APIs with strong reliability practices (typed contracts, validation/fallbacks, safe rollouts) and experience with queue-based microservice patterns (idempotency, retries, DLQs, monitoring).

Experience

Education

Key Strengths

• Owned and shipped a mission-critical AV automation + dashboarding product used daily for hundreds of live events
• Balances speed and quality via minimal reliable core, then rapid user-feedback-driven iteration
• Designs for reliability in live/real-time operations (validation, fallbacks, safe rollout practices)
• Strong full-stack architecture decisions (typed API contracts, backend-centralized transformations, single source of truth)
• Microservices/async messaging reliability at scale (idempotency, retries, dead-letter handling, monitoring)
• Builds internal tools that achieve fast adoption by reducing daily operational friction
• Took an LLM prototype to production by grounding outputs in real data (RAG) and enforcing structured JSON responses
• Improved reliability by diagnosing pipeline stages (input/retrieval/reasoning/output) and using logging + prompt version tracking
• Resolved inaccurate estimates by identifying weak retrieval and improving chunking, domain-specific embeddings, and similarity thresholds
• Built stakeholder trust through explainable breakdowns and iterative validation against historical projects
• Effective technical communicator: delivers hands-on demos with live prompt tweaks and log walkthroughs
• Strong pre-sales partnership: aligns technical capabilities to customer pain points and customizes live demos with client data
• Built end-to-end OCR-based license plate recognition system (perception + ML + integration) at Volkswagen
• Improved OCR robustness under glare/shadows/motion blur/odd angles via preprocessing and plate isolation pipeline
• Achieved ~95% character recognition accuracy
• Real-time pipeline debugging and latency reduction through profiling and removing redundant preprocessing
• Distributed pub-sub communication design using schemas/message contracts (Kafka, Spark workflows)
• Modular, message-driven architecture mindset aligned with ROS node/topic patterns