Krishnapriyanka Ponnaganti

About

ML/AI engineer with hands-on experience shipping production computer vision and GenAI systems, including a fabric defect detection platform that combined vision models with agentic LLM workflows to reach 89% human-inspector agreement at 200 ms latency. Also built a RAG-based code QA tool for developers and emphasizes production monitoring, evaluation, caching, and reusable Python service design.

Key Strengths

• Owned end-to-end ML system from experimentation through deployment and monitoring
• Production computer vision expertise in defect detection
• Improved F1 score from 0.55 to 0.82 on defect classification
• Achieved 89% agreement with human inspectors
• Optimized inference latency to 200 ms
• Built hybrid vision plus LLM multi-agent pipeline
• Strong production monitoring with MLflow, CloudWatch, and SNS
• Iterative improvement using disagreement cases and production errors
• Shipped RAG-based code QA system grounded in source code
• Applied LLM quality evaluation using relevance, faithfulness, and answer-quality scoring
• Reduced hallucinations through structured outputs and retrieval constraints
• Designed reusable Python services with modular FastAPI architecture
• Improved team velocity via standardized reusable ML APIs for forecasting and segmentation
• Deployment and reliability focus through Docker, testing, and CI/CD staging

Reference Highlights

• strong end-to-end ownership
• takes responsibility for the whole system
• understands requirements well
• strong problem understanding
• thorough experimentation approach
• delivers systems that work at scale
• focuses on real metrics
• ties work to business value
• builds reliable systems
• builds maintainable systems
• sets up proper monitoring
• provides documentation
• optimizes systems after deployment
• balances accuracy with performance
• handles large data volumes efficiently
• solutions are used to solve real problems
• rigorous about quality
• iterates to improve accuracy
• tests edge cases
• ensures production readiness before deployment
• solid and maintainable Python code
• handles deployment well
• code is trusted and depended on by others
• processing hundreds of items daily in production
• saved over 100 hours of manual work through automation
• thinks through tradeoffs well
• balances latency, cost, observability, and reproducibility
• reduced response times significantly with caching
• cost-conscious and avoids over-engineering
• sets up strong alerting and logging
• documents work so others can rebuild systems
• delivers systems that work reliably in production

Experience

Education

Skills

Languages

Certifications

Publications

2 publications