Riddhi Ketan Shah

AI Engineer at Qlik

Standardized reusable LLM evaluation workflows for Qlik’s agentic platform using evaluation datasets, success metrics, and benchmarking with LangSmith and Langfuse, improving reliability and output quality for 30K+ users. Built an internal operations assistant leveraging MCP-based connectors, Grafana/Splunk telemetry, RAG retrieval, and Jira actions to automate SLO reporting and platform-support queries, reducing response and incident triage time by 80%. Developed and deployed a modular RAG pipeline with LangChain, hybrid retrieval, reranking, secure MCP connectors, and access-aware retrieval, improving enterprise knowledge-query relevance by 45%. Designed multimodal input validation in the LLM Gateway and Go microservices using AWS Bedrock Guardrails and custom validators, enabling secure request handling while reducing inference latency by 40%. Created reproducible foundation-model experimentation workflows with A/B testing, prompt evaluation, failure analysis, and trace-based revi

AI/ML Engineer at Qlik

Led improvements to Qlik Answers (AI Assistant) by enhancing LLM response accuracy for 30K+ global users. Integrated LangSmith and Arize Phoenix into the agentic system to boost observability and reliability; developed hybrid RAG pipelines with LangChain and OpenSearch, including a reranker to improve domain-specific query relevance; architected multimodal input support in LLM Gateway and microservices with security scanning, reducing inference latency; automated LLM experiments with A/B testing across models (GPT-4, Claude, Mistral) improving model selection efficiency; created a test-generation agent with GitHub Copilot to automate Python/Go test writing, achieving 100% coverage.

Machine Learning Engineer at Synechron|Client: American Express

Developed and deployed an anomaly-detection system for large-scale financial data using XGBoost, EDA, and feature engineering, reducing false positives by 35% for downstream risk-review workflows. Built scalable PySpark and Airflow data pipelines processing 200K+ customer records, enabling reproducible ML workflows, automated batch training, and versioned model releases. Applied SHAP interpretability and MLflow experiment tracking to explain model predictions, compare model versions, and improve transparency for stakeholders. Adapted domain-specific LLMs using Hugging Face and LoRA/PEFT, evaluating validation loss, overfitting behavior, and response quality to improve precision by 45% on proprietary data. Deployed ML and LLM services as containerized REST APIs, integrating versioned inference workflows with reusable service interfaces for downstream applications.

Machine Learning Engineer at Synechron

Architected an end-to-end ML pipeline for fraud detection with automated feature engineering, processing 10K+ daily transactions; developed an ensemble fraud detection model (Isolation Forest, LOF, XGBoost) achieving 82% precision with 30% reduction in false positives; implemented MLflow-based monitoring for model accuracy and data drift across 15+ features; fine-tuned LLMs (LoRA/QLoRA) on proprietary data and built RAG pipelines with vector search to boost domain-specific query precision by 45%.

Software Developer Intern at Cisco

Developed a Python integration layer between network simulation and routing-validation systems, improving test orchestration and reducing manual setup time by 30% across 500+ routing configurations. Improved validation throughput using Python multithreading and structured test workflows, enabling automated scale testing across 1,000+ simulated network topologies.

Software Development Engineer II Intern at Cisco

Built a Python integration layer connecting network simulation and routing systems, reducing test setup time by 30% and enabling automated validation of 500+ routing configurations; improved large-scale Lisp network simulation throughput by 40% through multithreaded Python, enabling parallel testing of 1000+ topologies.