Raoof
Naushad

Leading a 13+ engineer team building the AI Core of a wealth-management platform serving family offices and institutions across the GCC.

• Architected a secure LLM Gateway with PII/SPII anonymization, prompt-injection defense, and policy-driven model routing based on invoked tools and data classification.
• Built a Claude-style agentic assistant operating entirely over private, internal MCP tools — creating documents, dynamic dashboards, and scheduled tasks in a fully private environment.
• Implemented fine-grained RBAC across every agentic use case, scoping retrieval and tool access to each user's entitlements.
• Deployed open-source models (Qwen, Kimi, DeepSeek) for private VNET/on-prem inference, cutting external-provider dependency and cost.
• Owned platform cost, reliability, and security (VNETs, Private Endpoints, Key Vault) with IaC across dev → staging → prod.

Invented and led enterprise GenAI patterns for Personal Banking under strict security and risk governance.

• Invented and led a patent-pending GenAI Personal Statement generator producing narrative financial insights from client data.
• Implemented the SafeAI Adoption Process and an AI Onboarding Pattern, standardizing secure private-LLM inference across the enterprise.
• Mined 5M+ client records, enhancing LLMs with prompt engineering and RAG (chain-of-thought, RAPTOR) for a 23% faster pipeline.
• Built a Llama 3 8B function-calling PoC to replace an OpenAI-based app — a potential 96.3% cost-per-score reduction.

Built data & ML systems at scale before moving fully into AI architecture.

• Factspan: PySpark on Azure Databricks transforming 5M daily records; automated testing framework (spot award for innovation).
• UST Global: BERT intent/entity NLP automating an IVR system; AWS Step Functions + Lambda pipeline with a 60% faster data load (UStar Award).