I'm an OCI Generative AI Black Belt at Oracle, working across enterprise and government customers in the UAE and broader EMEA region. This repository is a collection of the demo applications, agentic systems, reference architectures, developer coaching sessions, and content I've built to showcase what's possible with Oracle Cloud Infrastructure's Generative AI platform.
The work here spans the full spectrum of enterprise AI use cases — from document intelligence and multimodal analysis, to agentic workflows, compliance tooling, and RAG architectures — built to be production-credible and customer-ready.
A document-extraction engine that checks its own work. It reads invoices, purchase orders, receipts and delivery notes with Cohere Command A Vision (via langchain-oci), then runs a self-correcting loop — validating every figure against deterministic, pure-Python business rules and re-reading only the fields that don't reconcile, pass after pass, until the document balances. The headline score is a fact audit, not the model's opinion, so demo and live land on identical numbers. Schema-driven (presets plus an in-app schema editor), with on-demand (Frankfurt) or Dedicated AI Cluster serving. Built as FastAPI + React with a streaming, pass-by-pass loop view.
An agentic bid-response application on the OCI Generative AI Responses API. The agent reads a government or enterprise tender (RFP) PDF end-to-end, extracts every individual requirement, scores each into a COMPLY / PARTIAL / GAP compliance matrix, produces a bid/no-bid recommendation with win themes mapped to the evaluation criteria, and drafts the proposal executive summary — in English or Arabic — before exporting a ready-to-edit Word response pack. Built for bid teams that need to triage more tenders without missing a disqualifying clause.
An executive-ready speech-intelligence asset for enterprise contact centers, built on OCI Generative AI and an imported Whisper Large V3 Turbo endpoint. It turns call audio — uploaded, recorded, or pasted — into accurate multilingual transcripts, then into a typed intelligence package: QA score, sentiment curve, churn risk, key moments, compliance findings, a coaching plan, and CRM-ready action items, with the whole path kept inside OCI. Built as FastAPI + React.
Turns a single sample business document into hundreds or thousands of realistic synthetic versions, so teams can build and test document pipelines without ever sharing real customer data. Built on Gemma 4 on a Dedicated AI Cluster (via langchain-oci) for layout extraction and data generation, with an optional image-rendering path (openai.gpt-image-1.5 on the OCI Images API) for scan-like outputs ideal for OCR and extraction training data. Everything runs inside your OCI tenancy.
An agentic application that automates legal document review using the OCI Responses API. The agent performs multi-step reasoning over contracts and legal documents to surface risks, obligations, and key clauses — reducing manual review time significantly. Built with the OpenAI SDK against OCI's Responses API endpoint.
A Streamlit-based compliance application that uses Oracle Document Understanding (DU) and Generative AI to analyze regulatory documents alongside internal policies, identify coverage gaps, and produce structured gap analysis outputs. Built for regulated industries where policy alignment is business-critical.
A retail-focused application demonstrating OCI's native image generation capabilities. Takes product descriptions and generates polished product imagery, showcasing OCI GenAI's multimodal model support via Model Import on Dedicated AI Clusters.
A multimodal Generative AI application that analyzes images from car accidents and auto-generates structured insurance reports. Combines OCI Vision and Generative AI to extract damage details, produce narrative summaries, and output structured reports ready for claims processing.
A conversational chatbot built for insurance claim intake. Guides users through the claims process via natural dialogue, extracts structured data, and produces structured outputs — demonstrating how GenAI can replace rigid form-based workflows in financial services.
A health intake assistant that collects free-text symptom descriptions and extracts key clinical parameters into a strict JSON schema using an LLM. When information is incomplete, the app asks targeted follow-up questions until the schema is satisfied. It then explains likely conditions and suggested next steps in plain language, with full conversational memory for follow-up questions. Includes configurable knobs for similarity thresholds, coverage preferences, and common-disease weighting.
An end-to-end document intelligence application that combines Oracle Document Understanding with Generative AI to extract, classify, and analyze enterprise documents. Demonstrates the full document lifecycle from ingestion to structured output.
A Generative AI application that analyzes frames from uploaded videos or images using LangChain orchestration and OCI Generative AI. Determines content suitability and generates structured scene analysis — applicable to content moderation, media, and security use cases. Available in two versions: a LangChain implementation and a direct OCI SDK implementation (decode-images-and-videos-with-oci-gen-ai).
A RAG-powered chatbot for querying complex, multi-format reports — financial, health, HR, and others. Uses a combination of Document Understanding and vision LLMs to handle both the text content and embedded visual elements (graphs, charts, illustrations), enabling truly comprehensive document Q&A.
An application demonstrating OCI Generative AI's image generation capabilities, producing images from text prompts via OCI's native GenAI service.
A CV screening application that compares candidate CVs against a defined set of role requirements using LangChain and OCI Generative AI, ranking and surfacing the best-fit candidates. Designed to accelerate shortlisting in high-volume recruitment workflows.
An application that generates structured news articles using OCI Generative AI — demonstrating templated content generation workflows applicable to media, communications, and marketing use cases.
A minimal, clean implementation of Retrieval-Augmented Generation on OCI — built as a practical starting point for teams looking to implement RAG without heavyweight orchestration frameworks.
A Generative AI application that extracts key clinical information from patient referral letters (.docx format) and structures it for expedited patient handling. Built to reduce administrative burden in healthcare settings.
A multimodal application that extracts all key elements from PDF invoices into structured CSV output, using OCI's vision-capable language models. Production-ready for finance and procurement automation.
Presented as part of Oracle's EMEA Black Belt developer coaching programme — public sessions covering advanced OCI GenAI topics.
- Discovering Multi-modal Models for Complex Documents (Feb 2026) — Deep dive into using Llama 4 Scout and Maverick for PDF and image data extraction, co-presented with Ilayda Temir.
- Agents with OCI GenAI — Calling Tools (May 2025) — What's New in Tech session covering function/tool calling with OCI Generative AI service, agentic patterns, and the Responses API.
- Automate the Document Life Cycle with OCI GenAI (Feb 2026) — Published Oracle blog covering end-to-end document automation architecture, from ingestion through extraction, classification, and action.