Nicolas Docolas

AI EngineerCurrent

• —Building AI microservices with Python, FastAPI, LangChain, and RAG pipelines for enterprise NLP applications.
• —Applying prompt engineering and LLM evaluation techniques to improve response accuracy and reliability in production.

Machine Learning Engineer

• —Worked on the design and development of large-scale enterprise AI systems, focusing on agentic architectures, multi-agent orchestration, and production-grade LLM applications. Built and maintained scalable backend services using Python, FastAPI, LangChain, Docker, and CI/CD pipelines, emphasising clean architecture, modularity, and maintainability. Designed secure access control through Role-Based Access Control (RBAC) integrated at the application's data layer.
• —Developed orchestration layers for AI systems capable of dynamically routing tasks across specialised agents. Designed execution frameworks that decompose requests into Directed Acyclic Graphs (DAGs) of interdependent tasks, supporting parallel execution, dependency resolution, and pause-and-resume capabilities. Integrated Redis for persistent state management, improving robustness, reducing latency, and optimising token usage.
• —Applied advanced LLM engineering techniques including prompt engineering, Retrieval-Augmented Generation (RAG), and guardrails such as hallucination mitigation and controlled retrieval. Built pipelines enabling context propagation across tools and agents, ensuring reliable outputs in enterprise environments.
• —Contributed to conversational AI systems with dynamic agent routing, tool-based execution, and observability through logging and metrics. Developed capabilities such as translation, summarisation, document generation, and image generation using structured prompts and API integrations.
• —Engineered on-premises AI infrastructure by migrating cloud-based systems to multi-node, multi-GPU environments. Containerised applications with Docker, orchestrated with Kubernetes. Benchmarked and optimised open-source models (Qwen, Llama, DeepSeek) using quantisation techniques to balance performance, accuracy, and efficiency while improving privacy and reducing external dependencies.