.png)
.webp)
Corby is a utility construction and engineering services company based in Michigan, employing 600 to 700 staff as the largest of nine companies under the CMG parent group. Their data was scattered across Power BI and SharePoint, and their internal IT team was at full capacity.
Corby came to NineTwoThree to build an enterprise AI chatbot that gives employees direct access to company data and lets project managers query their own data without waiting on IT. They also needed a partner to help their newly hired ML engineer build production-ready architecture. This focus on internal AI efficiency over external flash is where companies see the most lasting returns.
Central knowledge base pulling from Power BI and SharePoint for instant employee access to company information.
General chat for all employees with permission-filtered responses (i.e. identity, role, union status, and tenure).
Routing agents that interpret questions, select the right data sources, and retrieve answers across 15+ Power BI tables.
Natural language querying across 15+ Power BI tables, converting questions into DAX queries with results matching manual reports.
Before building the system, Corby had to work through several obstacles that made off-the-shelf tools insufficient.
Corby's data lived in Power BI and SharePoint. Employees emailed HR for answers; project managers waited on IT for reports. With IT at full capacity, Corby needed a partner to build the system and train their new ML engineer.
The system had to answer the same question differently based on who asked. Permissions spanned three roles across nine CMG companies, plus filters for union status, location, and tenure. A single PTO question required checking company, union status, tenure, and location before responding.
Three factors drove the decision to build custom: ownership and maintenance (off-the-shelf tools create vendor lock-in), permission complexity (the multi-layered structure needed custom routing logic), and data complexity (most institutional knowledge lived in employees' heads, not documented schemas).
NineTwoThree built an AI knowledge base with two distinct user experiences, backed by a custom routing architecture that queries multiple data sources based on user identity and permissions.
A three-step workflow: Microsoft Entra verifies the user's company and role, a custom agent selects the right data sources, and the system returns filtered answers from Power BI and SharePoint. A union employee asking about 401k eligibility gets a different answer than a non-union employee.
3
Data sources (Microsoft Entra, Power BI, SharePoint)
Single-question chat for all employees to reduce HR emails and calls. Each query is routed to the correct data sources and filtered by user permissions, a core capability of well-built enterprise chatbots.
A Notebook LM-style chat with persistent history for project managers. Users ask natural language questions across 15+ Power BI tables; the system converts them to DAX queries. In a live demo, AI-generated results matched manual Power BI reports exactly.
NineTwoThree connected Langfuse and customized it so Corby's team can monitor the system and make informed decisions. We also trained Corby's ML engineer to maintain the system. The architecture enables expansion into timesheets, job management, and billing.
700
employees across 9 companies required a scalable knowledge transfer for frictionless adoption
Corby is targeting a 75% reduction in HR emails and calls and IT report requests by giving employees direct access to company information through construction AI tools.
Company managers query project data in natural language and get answers in seconds instead of waiting for IT reports.
Corby is redirecting time from repetitive operational tasks toward revenue-generating work, aiming to scale revenue without scaling costs, one of the strongest cases for internal AI.
The enterprise AI assistant architecture unlocks expansion into timesheets, job management, and billing, becoming the foundation for a broader enterprise AI strategy.
Build a new platform, or add AI capabilities to existing software, we deliver solutions that work in real operational environments.
Meet with founders Andrew Amann & Pavel Kirillov
Custom AI implementation roadmap
No sales pressure
