Sherry

About Sherry

From REST APIs to AI: My journey began 8 years in backend engineering focused on Python and microservices, and cloud-native technologies. Today, I develop and implement enterprise scale AI systems to solve genuine challenges. I transition from traditional software engineering (REST APIs, OOP, SOLID principles) to design-autonomous workflows AI/ML pipelines, enabling ethical, scalable AI integration.Core Expertise→ Generative AI & LLMs:Constructing retrieval-augmented generation (RAG) pipelines for LLMs like GPT and fine-tuning them into use-case specific solutions. Deploying vector databases such as Pinecone and MongoDB Atlas.Tools: LangChain, Hugging Face, PyTorch, OpenAI API.→ MLOps and AI infrastructures:Allied containerized workloads through Docker/Kubernetes on AI model orchestrators like AWS SageMaker and MLflow for scalable operations.Implemented automated CI/CD frameworks for AI task pipelines leveraging GitHub Actions and Jenkins.→ Native AI cloud:Implemented cost-effective data lake and model training infrastructures using AWS Bedrock and Lambda (serverless inference) and S3.Developed AI enabled FastAPI and GraphQL APIs with sub 100ms response time.→ Leading Engineering Data AI:Designed AI model ETA pipelines using PySpark in conjunction with AWS Glue and Snowflake.Specialist in vector embeddings and structuring unstructured data with Pandas and NumPy.→ Old world AI modernization:Transformed monolithic applications to AI infused microservices, adding components such as chatbots and fraud detection capability systems.

Experience

1. Senior Ai Ml Engineer

   Crum & Forster

   Aug 2024 — Present · NY, US

   Built a RAG pipeline using LangChain and FAISS to summarize 10K+ unstructured claims/month, reducing adjuster review time by 40%.‣ Deployed a model-chaining workflow (XGBoost → PyTorch) for real-time risk scoring, achieving 90% precision in fraud detection.‣ Engineered <200ms latency with Apache Flink streaming and SageMaker endpoints, hosted on hybrid cloud (AWS EC2 + Kubernetes).Integrated SHAP/LIME for explainable AI and Neo4j knowledge graphs to map insurance domain logic.‣ Designed scalable web applications using Django/Flask, integrating 15+ relational (PostgreSQL, MySQL) and NoSQL (MongoDB) data sources.‣ Built RESTful APIs for internal/client-facing modules, ensuring compliance with security protocols (data segregation, GDPR).Architected hybrid AWS cloud storage solutions (S3, DynamoDB, OpenSearch) and deployed serverless workflows (Lambda, Fargate).‣ Automated testing (Pytest/PyUnit) and CI/CD pipelines (Bitbucket) to reduce deployment errors by 40%.Tech Stack: Llama-3, Docker, Kafka, TensorFlow, AWS SageMaker, Azure Purview, Python, Flask, REST APIImpact: Cut operational risks by 35% while processing claims 5x faster.