Work Highlights

Enterprise NL2SQL Agent

• Led the development of SOMA-SQL, a generalizable enterprise NL2SQL system that resolves ambiguity across user intent, schema context, and execution feedback through planning, probing, and iterative refinement.
• Built production agentic NL2SQL systems for robust, cross-dialect SQL execution across enterprise workflows.
• Achieved #1 results on bilingual NL2SQL benchmarks, including Archer and Spider 2.0 Lite.

Proactive Insight Discovery for DevOps

• Developing a proactive insight layer for SRE agents that correlates operational signals and service dependencies to surface emerging incidents before formal ticket creation.
• Helps on-call engineers triage and resolve tickets faster by identifying likely duplicate or downstream dependency incidents, reusing relevant investigation context, and estimating potential impact.
• Extends reactive incident analysis with evidence-grounded operational reasoning to support more reliable root-cause investigation and incident response.

• Led the development of a streamlined, reproducible LLM evaluation harness across multiple domains (Agent, Multilingual, NL2SQL, Code, Responsible AI, Internal), with standardized logging/tracking and confidence-aware reporting for leadership model-release decisions.
• Built Oracle Fusion Agent evaluation frameworks for tool-level and end-to-end benchmarking of internal employee-assistance use cases (for example, benefits analysis and account advisory).
• Built open-source and in-house RAG benchmarking pipelines and evaluation plans for RAG/Agent services, delivered multimodal RAG PoCs (vision embeddings + multimodal models), curated benchmark datasets, and supported partner RAG evaluation for product integration decisions.

• Led the team to deliver synthetic data generation across text, image, and audio for internal ML service fine-tuning and evaluation, and drove customer-facing PoCs across OCI teams (Language, Speech, GenAI, CX, Utilities), enabling broad internal adoption and substantial cost savings.
• Delivered SDG Service V1 and built the end-to-end SDG platform (libraries, config/evaluation pipelines, and APIs) to improve scalability, quality, and reproducibility.
• Led collaboration with multilingual linguists to establish human-in-the-loop (HIL) synthetic data validation and review workflows.

• Led development, multiple releases, and launch of OCI anomaly detection production service, including deep-learning capabilities for pure univariate, univariate+categorical, and discrete univariate signals.
• Built core preprocessing, training, and inference pipelines that became shared service infrastructure for extensibility and faster iteration across kernels.
• Built a graph-based anomaly detection and forecasting solution for Danish power plants on OCI Anomaly Detection, enabling earlier breakdown alerts and ~30% higher operational efficiency.
• Prototyped non-time-series anomaly detection for financial transaction and SaaS report analysis using dimensionality reduction and reconstruction-based modeling.

• Led healthcare AI projects end-to-end, from solution design to production deployment, for healthcare-sector customers across insurance, legal, and compliance functions.
• Delivered pre-authorization automation for bariatric, infertility, and spine surgery review for a major U.S. health insurance company, improving turnaround speed and generating multi-million-dollar annual savings.
• Built NLP and anomaly-detection systems with scalable cloud architectures and analytics dashboards for multiple healthcare corporations, including medical-device companies, to support operational decision-making.