GROUP AI PILOT · LEGACY MODERNIZATION
Multi-agent code migration: RPG to Java in minutes
The problem: modernizing legacy code is slow, risky and manual.
In this client's case, a team of more than five people needed over two sprints to migrate a single RPG
component, below 80% effectiveness, with API documentation written by hand.
The solution: a multi-agent AI migration tool that transforms
legacy code into modern systems: one engineer runs a whole project, with several components and their
dependencies, and gets automated API documentation and Azure DevOps integration.
The impact: a full project migrates in under 10 minutes at 90 to
95% effectiveness, with manual effort, risk and cost cut accordingly.
How it was built: LangChain and LangGraph orchestrate
specialized agents that analyze the legacy code, generate the modern equivalent and document the APIs, with
embeddings and analysis results stored in MongoDB. Python backend, Streamlit and Angular front ends, all
deployed on AWS with Terraform and Docker and wired into Azure DevOps for repositories and work items.
Multi-agent
LangGraph
90-95% effectiveness
2+ sprints → minutes
GROUP AI PILOT · TALENT OPERATIONS
Multi-agent talent profiling and assignment
The problem: matching people to open roles took an average of
three days and three people per search, with profile accuracy of 70 to 75% and résumés updated by hand.
The solution: a multi-agent LLM system that profiles candidates,
keeps their records current automatically and proposes the best assignment for each vacancy.
The impact: searches resolve in minutes with one person involved,
profile accuracy rose to 92%, and more of the candidates put forward end up selected.
How it was built: a LangGraph agent workflow reads
structured talent data from PostgreSQL, stores embeddings and LLM analysis in MongoDB, and surfaces results
through Streamlit and Angular front ends. Deployed on AWS with Terraform and Docker, with MCP tools for
database access.
LLM agents
3 days → minutes
92% profile accuracy
GROUP AI PILOT · SRE OPERATIONS
AI assistant for incident impact analysis
The problem: SRE engineers spent much of their day on manual
impact analyses and post-incident work: two analyses a day, four hours to respond to an incident, and three
to four hours to produce root cause, action plans and lessons learned.
The solution: a multi-agent assistant on AWS Bedrock that
automates validations and controls, retrieves similar past incidents and drafts standardized documentation
from prior experience.
The impact: five analyses a day, incident response and
related-incident search in about five minutes each, root cause and action plans in minutes, and consistent
documentation on every incident.
How it was built: agents run on AWS Bedrock orchestrated
with LangChain and LangGraph, retrieve similar past incidents through embeddings stored in MongoDB, and
generate standardized reports. Infrastructure is provisioned with Terraform and shipped in Docker containers
on AWS.
AWS Bedrock
4 h → 5 min response
2.5x daily throughput
GROUP AI PLATFORM · KNOWLEDGE ASSISTANTS
Knowledge-base chatbots, configured in minutes
The problem: every team wants an assistant that answers from its
own content, but each one usually means a new engineering project.
The solution: a chatbot platform on AWS Bedrock and Claude where
non-technical users load text, PDFs, video and images into a knowledge base, with vector search, topic
guardrails and PII protection built in.
The impact: a working assistant configured in minutes instead of
weeks, with encrypted data, moderated answers and cloud costs kept in check.
How it was built: a serverless stack on AWS Lambda and
API Gateway with S3 for content uploads. Amazon Bedrock provides Claude, embeddings, knowledge bases and
guardrails; Amazon Transcribe and Rekognition ingest audio and video; MongoDB Atlas handles vector and
full-text search.
Claude on Bedrock
RAG
Minutes to configure
