• Designed and implemented a production-grade Generative AI RAG platform for BPX Energy, delivering secure, low-hallucination Q&A over enterprise data with sub-second to low–single-digit-second response latency for most user queries.

• Built a semantic retrieval layer using Amazon OpenSearch as the vector store, supporting thousands of indexed documents and embeddings with consistent retrieval performance under concurrent user load.

• Integrated LLM inference via Amazon Bedrock using Claude 3.5 Haiku, optimizing for low-latency responses and cost-efficient inference compared to larger foundation models.

• Implemented event-driven ingestion and indexing pipelines using AWS Lambda and Amazon EventBridge, enabling near-real-time document updates while keeping ingestion costs fully serverless and usage-based.

• Designed prompt templates and semantic routing logic to dynamically assemble prompts from user input, retrieved context, and session state, reducing hallucinations and improving answer relevance across multi-turn conversations.

• Built session and conversation memory management using Amazon DynamoDB, supporting high-concurrency access patterns with predictable single-digit millisecond read/write performance.

• Enforced least-privilege security controls using AWS IAM for Bedrock, Lambda, OpenSearch, and DynamoDB, ensuring secure agent execution in a regulated enterprise environment.

• Implemented end-to-end observability with Amazon CloudWatch, tracking request latency, error rates, and invocation counts to support operational stability and rapid troubleshooting.

• Architected the platform for horizontal scalability, allowing independent scaling of ingestion, retrieval, inference, and session layers to handle growth in users, documents, and query volume without redesign.

• Optimized overall system cost by combining serverless ingestion, lightweight LLM models (Haiku), and semantic retrieval, keeping per-query costs low while maintaining enterprise-grade accuracy.

• Architected and led the development of production-grade GenAI systems using Retrieval-Augmented Generation (RAG) and Text-to-SQL to deliver explainable, guideline-grounded insights.

• Built LLM-powered APIs and chatbots using FastAPI and managed inference platforms (AWS Bedrock, Vertex AI).

• Designed hybrid retrieval workflows combining vector search and deterministic SQL, reducing hallucinations by >90% and improving response relevance by 35%.

• Implemented Lang Chain-based RAG pipelines with OpenSearch vector indexing to support semantic retrieval over large clinical document corpora.

• Developed secure Text-to-SQL execution layers enabling natural-language analytics over Snowflake with strict read-only enforcement.

• Integrated observability and drift monitoring using CloudWatch to track LLM latency, retrieval quality, and system health.

• Partnered with clinicians, data teams, and leadership to deliver role-aware GenAI experiences for physicians versus medical directors.

• Detected and contained an active AWS security breach caused by exposed credentials in source control.

• Performed deep IAM investigation and identified attacker-created IAM users, access keys, and privilege escalation paths used for persistence.

• Restored all production and non-production environments within 24 hours.

• Led post-incident security redesign using AWS SSO, Azure AD integration, least-privilege IAM, Prowler.

• Sole owner of AWS and GCP cloud platforms, responsible for infrastructure, IAM, CI/CD, monitoring, and disaster recovery.

• Implemented AWS Organizations and GCP organizational structures with proper account, project, and access isolation.

• Designed and implemented end-to-end disaster recovery (DR) plans.

• Implemented continuous security posture management using Prowler, AWS Config, and Security Hub.

• Designed and enforced RBAC models, permission sets, and MFA-based access using IAM Identity Center and Azure AD.

• Developed Python-based automation scripts to support healthcare data processing, system monitoring, and operational workflows.

• Worked with structured data formats (CSV, XML, JSON) using Python to clean, transform, and prepare datasets for downstream systems.

• Integrated Python scripts with Linux environments to schedule and execute recurring jobs using cron.

• Assisted in debugging, testing, and improving Python applications, following clean coding and basic performance optimization practices

• Migrated on-prem workloads to AWS and Azure using Terraform and ARM templates.

• Automated deployments via Ansible integrated with Jenkins and GitHub.

• Created reusable Terraform modules and IaC templates for repeatable multi-region deployments.

• Installed, configured, and maintained Jenkins/Hudson for continuous integration and end-to-end automation of builds and deployments.

• Provided system administration support for over 100 servers across diverse platforms and operating systems.

• Developed and documented software release management procedures, including release notes for scheduled releases.

• Managed Linux environments, deploying web applications using Puppet and automating processes with BASH and Shell scripts.