Achref Hamdi

Myra++ (Version 1): – Developed a Retrieval-Augmented Generation (RAG) system leveraging company data (Jira tickets, wiki documentation) to enhance support team efficiency by optimizing search for recurring issues. – Collected, visualized, and analyzed raw data; performed cleaning and transformation for indexing into a vector database. – Designed and implemented a retriever class and system prompt for accurate response generation. – Deployed the solution as a SaaS API using Docker and Kubernetes; automated data workflows with a Kubernetes job. – Monitored system performance to ensure reliability and scalability. – Technologies: Python, RAG, LlamaIndex, Qdrant, Docker, Kubernetes, GitLab, ArgoCD, LangFuse. Smart Mapper: – Built a RAG-based system to provide users with mapping solutions (e.g., mapping formulas, converter code, global variables) for transforming elements between contexts (e.g., MT 502 to MT 515). – Collaborated with clients to understand business requirements; collected, visualized, and cleaned relevant data. – Transformed and indexed data into a vector database; developed a retriever class and system prompt for precise mappings. – Delivered the solution as a SaaS API with continuous monitoring for performance optimization. – Technologies: Python, RAG, Qdrant, LlamaIndex, Docker, Kubernetes, GitLab, SaaS, REST API. Myra++ (Version 2): – Enhanced Myra++ (Version 1) into a multi-agentic GraphRAG solution using user feedback, improving intelligence and response accuracy. – Transformed and indexed data into a Neo4j graph database; designed agents (orchestrator, decision controller, Jira agent, wiki agent). – Built retriever and index classes for efficient data retrieval and processing. – Deployed the solution as a SaaS API with ongoing monitoring for stability and performance. – Technologies: Python, Neo4j, Agentic RAG, LangChain, LlamaIndex, GraphRag, Docker, Kubernetes, ArgoCD. EasyAgreement: – Developed an AI-driven collateral management solution that digitizes legal agreements and automatically extracts structured eligibility rules (asset type, haircut, rating criteria, remaining maturity, eligibility conditions, ...). – Achieved 96% extraction accuracy, significantly improving operational efficiency by converting complex PDF agreements into machine-readable rule sets. – Designed and implemented a complete document-processing pipeline combining OCR, layout analysis, and multi-agent reasoning to extract rules with high accuracy. – Built specialized extraction agents (asset type, haircut, maturity, rating criteria, eligibility) coordinated by a workflow-of-thought orchestration system for reasoning, verification, and conflict resolution. – Engineered a multitenant SaaS API with per-request isolation using Kubernetes namespaces, secure JWT-based tenant identification, and strict data separation. – Containerized and deployed the full solution using GitLab CI/CD, Docker, Kubernetes, and ArgoCD with automated monitoring and versioning. – Technologies: Python, OCR, Multi-Agent AI, RAG, Qdrant, FastAPI, Docker, Kubernetes, GitLab, ArgoCD, SaaS Architecture.

– Develop and deploy models for person detection, recognition, and classification by age and gender in commercial spaces – Keywords: Deepstream, Triton, TensorRT, TAO Toolkit, Jetson nano. FastAPI.

– Developed and deployed models for real-time person detection from video streams. – Keywords: DeepStream, Triton, Jetson nano, TensoRT.