Back to projects

SAS-to-Dataiku Conversion AI Agent

An LLM-powered agent that converts SAS scripts into Dataiku workflows to accelerate platform migration.

Application sectors

  • Banking
  • Platform Migration
  • Data Engineering

Technologies

  • GenAI
  • LangGraph
  • SAS
  • Dataiku
  • Python
Developer working on code migration across multiple screens

Business issue

Why I was brought into the project

Organizations moving away from SAS need to convert complex scripts into modern analytics workflows without spending months on manual rewriting.

Context

The environment around the project

SAS migration programs require a large amount of manual analysis and rewriting.

Functional environment

The functional context centered on reducing migration time, controlling costs and helping teams move legacy analytics assets to Dataiku.

Technical environment

The technical environment combined LangGraph, LLM orchestration, SAS script analysis, Dataiku workflow generation and configurable model selection.

Challenges

The system had to understand legacy SAS logic, generate usable Dataiku workflow structures and stay configurable across different migration contexts.

Solution

My contribution and its impact

My contribution to the project

I developed an intelligent conversion agent with a configurable and LLM-agnostic architecture, enabling users to choose models, manage cost and streamline processing time.

  • LangGraph-based conversion agent
  • LLM-agnostic configuration layer
  • SAS-to-Dataiku workflow generation
  • Validation process with technical and business stakeholders

Impact

The agent helped accelerate migration work by automating a large part of the conversion process while keeping expert validation in the loop.

  • Reduced migration complexity
  • Faster first conversion of SAS assets
  • More controlled use of LLMs across migration scenarios

Impact metrics

  • Automated conversion
  • LLM-agnostic design
  • Migration acceleration

Approach

How I structured the work

  1. Analyze SAS scripts and identify reusable transformation patterns.
  2. Design a LangGraph workflow to orchestrate conversion, validation and output generation.
  3. Collaborate with stakeholders to review generated Dataiku workflows and improve conversion quality.

Takeaways

What I learned from this project

  1. Legacy migration needs structured reasoning, not only code generation.
  2. Configurability is essential when different clients have different cost, quality and model constraints.
  3. AI-generated migration outputs must be validated with both technical and business stakeholders.