Hadi amiri

Delivered client-facing generative AI and cloud data platforms across telecom, federal government, healthcare, and energy. Optimized an air-gapped, Kubernetes-hosted agentic PDF pipeline using vision-LLM enrichment and addressed a cross-encoder reloading bottleneck, reducing ingestion time 2.3x and RAG agent response time from 36s to 6.6s (~80%). Built hybrid dense+sparse retrieval on Qdrant with content-hash deduplication and extended ingestion to additional file formats; implemented synthetic golden datasets and RAG evaluation using Langfuse tracing. Benchmarked 11 vision LLMs for production selection and built LangGraph agent tooling with parallel validation/judge nodes and bounded auto-fix loops, keeping LLM usage deterministic-first and human-gated inside the client’s Azure AI Foundry tenant. Deployed a production PII de-identification service (Microsoft Presidio + RoBERTa NER, optional LLM layer) reaching 94.9% redaction accuracy and 0% over-redaction after diagnosing systemic

Built production RAG ingestion and retrieval pipelines for enterprise clients, including hybrid dense+sparse retrieval with Qdrant, vision-LLM enrichment, chunking, cross-encoder reranking, and content-hash deduplication. Implemented RAG fallback strategies leveraging Cosmos DB. Developed agentic systems using LangGraph state machines for multi-skill tool orchestration, provider-agnostic LLM clients (Anthropic, Azure OpenAI, Azure AI Foundry), MCP tool integration, and human-in-the-loop gating. Created LLM evaluation harnesses with synthetic golden datasets using RAGAS and DeepEval, and added observability via Langfuse (datasets, tracing, AgentOps) on production Kubernetes. Delivered cloud data engineering capabilities including SSIS-to-Fabric/Databricks migration automation, medallion (bronze/silver/gold) architecture, Unity Catalog and Delta Live Tables, data masking, and event-driven schema-change detection. Deployed and governed AI services on Azure ML managed endpoints with Docker

Diagnosed and resolved telecommunications and network system issues, supporting configuration, maintenance, and upgrades for enterprise environments.