Practical AI integration
AI connected to the knowledge and workflows that matter.
I’m Ahmad Sadieh (s3dyeh). I integrate large language models, retrieval-augmented generation and AI agents into existing products, connecting the AI feature to the data and systems it needs to be useful.
What is retrieval-augmented generation?
Retrieval-augmented generation, or RAG, retrieves relevant information from a knowledge source and gives it to a language model as context. It is useful for assistants that need to answer questions about documents or business information.
A RAG system needs more than a model connection. The retrieved material must be relevant, access must match the user’s permissions, and answers need evaluation. Retrieval can improve the context available to a model, but it does not guarantee that every generated answer is correct.
LLM integration and AI agents
A knowledge assistant answers questions; an agent can also coordinate tools and steps in a workflow. I work on connecting these capabilities to application APIs and enterprise systems. The design depends on the task, the available data and the actions the system is allowed to take.
- Knowledge-based assistants grounded in relevant information.
- AI workflows integrated with existing application behavior.
- Clear access boundaries and evaluation of the resulting output.
FastAPI, LangChain, OpenRouter and Ragas
My AI toolkit includes FastAPI for Python APIs, LangChain for retrieval and agent workflows, OpenRouter for model access, and Ragas for RAG evaluation. I choose the pieces around the application’s requirements and its existing infrastructure.
Useful evaluation starts with the questions and tasks the application needs to handle. Retrieval quality and generated-answer quality are separate things to check, especially when the assistant uses internal knowledge.
Internal AI and data-quality integration
I implemented the backend and full AI integration for an internal AI and data-quality platform. The scope included anomaly, bias, duplication and sentiment detection alongside a configurable survey and rule builder.
The platform also connected an AI RAG system with reporting and monitoring workflows. My portfolio describes the engineering scope of this internal system without exposing its application or private data.