Projects
Selected Projects
OCI Generative AI Safety
Research and model development for Oracle's enterprise AI safety stack — the guardrails, content moderation, and policy-enforcement systems that gate generative AI in regulated and multilingual production environments.
The core technical challenge is translating broad safety, privacy, and policy requirements into measurable, testable model behavior. The work spans guard-model training and red-teaming, multilingual content moderation, jailbreak and prompt-injection evaluation, PII detection, and deployment-time risk analysis. It connects LLM alignment research to the operational constraints of enterprise deployments — latency, cost, observability, and policy auditability.
Public references
- Oracle Documentation: Guardrails for OCI Generative AI
- Oracle Release Notes: OCI Generative AI supports AI Guardrails for on-demand mode
Oracle Code Assist
Research and modeling work behind Oracle Code Assist, Oracle's enterprise AI coding assistant. The work focused on code-intelligence capabilities for real engineering environments — repository-aware completion, code explanation, test generation, and code modernization across Java, SQL, and SuiteScript.
The technical thread runs from continued pretraining on internal code corpora, through preference-based post-training using compiler feedback, static-analysis signals, and developer accept/reject telemetry, to a CWE-aligned secure-code evaluation harness used for release gating. The deployed system reaches tens of thousands of Oracle developers.
Public references
- Oracle Product Page: Oracle Code Assist
- Oracle Blog: Oracle Code Assist: AI companion to boost developer velocity
- Oracle Newsroom: Code Assist beta announcement, CloudWorld 2024
Projects Previously Led
OCI Vision and Document Intelligence
Technical lead for Oracle's vision and document AI stack — the production models behind OCI Vision and the document-understanding services used inside Oracle Cloud.
The work spanned image classification and object detection, OCR, layout-aware document understanding, table structure recognition, and key-value extraction from unstructured documents. Models were trained at cloud scale on synthetic and real document corpora and shipped behind enterprise APIs supporting procurement, finance, logistics, and ERP automation. The research thread connects computer vision, multimodal foundation models, and information extraction to the deployment realities of high-throughput enterprise pipelines.
Public references
- Oracle Product Page: OCI Vision
- Oracle Newsroom: Oracle launches OCI AI Services suite
- Oracle Blog: Introducing Oracle AI Services