Intership on Human-AI Systems for Logic and Code
Pubblicato il 17-08-2026 - Consulting / Consulenza / Coaching in Milano
Company
SDG Group . We are a consulting firm specialized in AI, Data & Analytics. Business agility is the ability of an organization to adapt quickly to market changes, both internally and externally. This cannot exist without becoming a truly data-driven company. SDG Group achieves this by co-creating optimal solutions with its customers, leveraging Data & Analytics services through a unique combination of business domain expertise and state-of-the-art technologies delivered by industry-leading talent. We are pioneers in AI, Data & Analytics consulting, and we are committed to unlocking organizations' hidden potential by offering in-depth analytics expertise. SDG is a global company located in multiple countries with over 2700 employees and more than 700 customers. Project Title
Implementing Human-in-the-Loop Agentic Systems for Automated Logic Discovery and Optimized Code Generation.
Project Description
The candidate will join the development team of AIMi , a modular framework that utilizes a multi-agent generative AI system to streamline the end-to-end (E2E)
migration of legacy code to modern platforms.
Technical Context: The platform is built on a Python/FastAPI backend using LangGraph to manage interconnected agents via a graph-based workflow. It employs a "Human-in-the-Loop" approach to ensure expert supervision during the discovery and generation phases. The system architecture is designed to be model-agnostic, integrating with LLMs such as Google Gemini, OpenAI, and Claude.
The Student's Role: Working under the guidance of a Senior Architect , the student will contribute to the design and implementation of new specialized agents, specifically focusing on:
Discovery Agents: Tasked with tracing functional dependencies and analyzing the logic of existing legacy systems (e.g., SAS).
Conversion/Generation Agents: Responsible for translating business logic into "Functional Recipes" and subsequently into optimized target code (e.g., dbt models).
This project offers