Rajesh Sudam

Senior AI/ML Engineer with 10+ years of experience designing and deploying production AI systems, including Generative AI, LLM-powered applications, agentic workflows, and responsible AI solutions. Experienced in applied machine learning research, model evaluation, human-in-the-loop AI systems, AI governance, and scalable ML platforms delivering enterprise-grade user experiences. Skilled in building large-scale image/video data pipelines, implementing comprehensive evaluation frameworks including quantitative metrics, ablation studies, and hyper-parameter optimization.

Hands-on with multi-GPU and distributed training workflows using PyTorch. Also brings deep expertise in LLM integration, RAG pipelines, and scalable cloud-native architectures on AWS. Proven track record delivering measurable outcomes across banking, enterprise, and scientific research environments.

Senior AI/ML Engineer at PNC Bank (2024-10 – Present)

• Architected and deployed LLM-augmented banking workflows integrating GPT-4/Claude APIs with Spring Boot microservices, enabling AI-powered document processing, anomaly detection, and intelligent customer transaction summarization at enterprise scale.
• Built internal Retrieval-Augmented Generation pipelines using LangChain, pgvector, and AWS Bedrock to surface context-aware insights from unstructured banking compliance documents, reducing manual review time by 35%.
• Designed and implemented agentic document-processing workflows using LangGraph and LangChain, enabling automated document classification, context retrieval, compliance summarization, API orchestration, and human-in-the-loop approval processes.
• Designed offline evaluation frameworks for LLM applications using quantitative metrics, prompt benchmarking, A/B testing, and human-in-the-loop feedback to optimize model quality, reduce hallucinations, and improve business outcomes.
• Implemented OpenAI GPT-4 and Claude function calling workflows with structured JSON outputs, enabling reliable integration of LLM agents with internal banking APIs, compliance systems, and document-processing services while reducing response parsing errors and improving workflow automation.
• Developed enterprise AI applications leveraging GPT-based models for document summarization, compliance analysis, and conversational assistants.
• Orchestrated multi-step AI workflows using AWS Step Functions and AWS Bedrock, enabling scalable model inference and automated retraining pipelines.
• Deployed and monitored ML models on Amazon SageMaker using managed endpoints, automated hyperparameter tuning, and model versioning.
• Defined infrastructure using CloudFormation and Terraform templates to provision secure, repeatable AWS environments.
• Engineered resilient LangGraph-based agent workflows incorporating conversational memory, tool routing, retry strategies, fallback handling, and human-in-the-loop escalation mechanisms, improving reliability, traceability, and task completion rates for enterprise AI applications.
• Developed LLM evaluation frameworks incorporating safety checks, hallucination monitoring, prompt guardrails, and responsible AI governance practices for enterprise-scale GenAI applications.
• Implemented MLflow-based model lifecycle management and Evidently AI monitoring pipelines to detect data drift, monitor model performance degradation, and trigger automated retraining workflows.
• Applied advanced prompt engineering techniques and custom AI instruction frameworks to accelerate developer velocity code generation, refactoring, and unit test creation via GitHub Copilot and GPT-4 improved team delivery speed by 30%.
• Built automated ML lifecycle pipelines using MLflow, Evidently AI, and AWS services to orchestrate model training, validation, deployment, drift detection, and retraining for enterprise AI applications.
• Engineered and automated end-to-end CI/CD pipelines via Jenkins and Git, cutting deployment time by 40%, implemented OAuth2-secured Spring Boot microservices sustaining 500K+ concurrent banking sessions with zero downtime.
• Architected scalable event-driven microservices using Java 17+, Spring Boot, Apache Kafka, and Redis caching, delivering high-throughput asynchronous processing for real-time banking transaction pipelines.
• Developed React-based operational AI dashboards integrated with Spring Boot microservices, enabling real-time monitoring of ML model outputs, transaction anomalies, and enterprise KPIs.
• Led AI/ML adoption initiatives: evaluated LLM and generative model architectures (LangChain, LlamaIndex, diffusion-based models), established GenAI and computer vision best practices, conducted model evaluation including ablation studies and hyper-parameter sweeps, and championed responsible AI governance across engineering teams, explored multi-GPU training strategies using PyTorch DDP for scaling model experimentation.
• Configured AWS services (EC2, RDS, Lambda, CloudWatch, Bedrock) for scalable AI model serving and infrastructure monitoring, implemented proactive alerting achieving 99.9%+ service availability for critical banking systems.
• Collaborated with security and compliance teams implementing Spring Security and OAuth2 across 15+ REST APIs, designed zero-trust microservices layer passing all regulatory security audits.

AI/ML Engineer at Siemens (2023-08 – 2024-10)

• Designed and integrated AI/ML inference services into distributed enterprise workflows, deploying model endpoints on Kubernetes with auto-scaling, achieving sub-100ms latency for real-time industrial analytics use cases.
• Built centralized feature engineering pipelines using Feast to serve online and offline features for predictive maintenance and industrial analytics models, ensuring training-serving consistency across environments.
• Developed XGBoost and LSTM models using industrial IoT telemetry data to predict equipment failures and optimize maintenance schedules, improving operational reliability across manufacturing systems.
• Implemented schema enforced structured output contracts for Kubernetes hosted ML inference endpoints, ensuring downstream industrial control systems received validated, well-formed predictions and rejecting malformed or out-of-distribution model responses.
• Designed, trained, tuned, and deployed XGBoost and LSTM models on 10M+ daily industrial IoT events, performing feature engineering, hyperparameter optimization, cross-validation, and model performance monitoring to improve predictive maintenance accuracy by 25%.
• Designed distributed data pipelines using Spark Streaming and Kafka to process and analyze 10M+ daily events, implementing scalable feature engineering and real-time inference workflows for production ML systems.
• Implemented automated model monitoring and performance tracking across Kubernetes-hosted inference services, ensuring reliabil