Ampere Financial, a U.S.-based innovator in energy transition financing, is reimagining how small and mid-sized industrial businesses decarbonize.
The challenge? Critical customer data, utility bills, equipment images, rebate rules, was scattered across PDFs, images, and spreadsheets, making it hard to turn into actionable intelligence.
The problem: missed rebate opportunities and complex energy guidance
Ampere Financial’s customer advisory process was slowed by the complexity of handling diverse data sources. Utility consumption records, equipment specifications, and rebate program details were spread across PDFs, images, and CSV files, making it labor-intensive to compile a complete view of a client’s eligibility for energy efficiency incentives. The lack of a unified data pipeline meant that rebate opportunities were often overlooked, leading to missed savings for clients and reduced adoption of electrification upgrades.
Additionally, traditional dashboards required manual navigation and interpretation, limiting personalization and forcing advisors to spend hours preparing reports. Ampere needed a secure, AWS-native agentic AI solution for decarbonization capable of ingesting this fragmented data, performing cross-table reasoning, and delivering accurate, location-specific guidance instantly, without the need for complex technical queries.
The solution: agentic AI for decarbonization with cross-table reasoning
To address these challenges, GoML and Ampere Financial co-developed a secure, AWS-first multi-agent AI architecture capable of turning raw, fragmented energy data into dynamic conversations.
Specialized AI agents were deployed to orchestrate a seamless, natural language copilot experience that Ampere employees can use to gain accurate insights about rebates.
The multi-agent solution leverages Amazon Bedrock (Claude Sonnet 3.7) for query generation, AWS Lambda for compute, S3 for storage, and a Supabase integration for structured data handling designed for scalability to nationwide deployment.
Rebate Supervisor Agent
This agent manages the flow of information from the user to different specialist agents and is responsible for assigning tasks to the right specialised Bedrock agents.
Document Ingestion Agents
These agents extract data from PDFs, images, spreadsheets, and other fragmented sources, converting unstructured energy data into structured formats usable by other agents.
Query Translation Agent
The agent translates user natural language queries into the correct structured SQL queries and synthesizes a contextual, accurate response aggregating data from other agents or sources.
Cross-Table Reasoning Agent
The Cross-Table Reasoning Agent integrates and reasons across diverse tables (equipment, utility consumption, rebate rules). These agents enable comprehensive, multi-source insights for downstream tasks.
Rebate Matching Agent
This agent identifies and matches applicable federal, state, and local rebates to customer scenarios based on equipment, usage, and ZIP code logic, surfacing all eligible opportunities dynamically.
Assurance Agent
The Assurance agent ensures outputs are accurate, hallucination-free, and compliant with regulatory boundaries before being surfaced to stakeholders (especially when dealing with financial or eligibility data).
Conversational Interface
The agent manages real-time user interactions, presenting results in a user-friendly, accessible format, and possibly routing queries to specialized agents as needed.
Architecture: Built with Amazon Bedrock AgentCore
The secure, AWS-first multi-agent AI architecture was built on Amazon Bedrock AgentCore leveraging the AgentCore Gateway and runtime.
The impact: smarter rebate usage, faster electrification decisions
Ampere Financial’s agentic AI decarbonization copilots delivered measurable results during the PoC phase:
- 65% faster rebate discovery: Instant ZIP code–based matching replaced manual lookups.
- 50% reduction in advisory preparation time: Conversations replaced dashboard navigation.
- 40% increase in identified rebate opportunities: Improved cross-referencing between equipment data and rebate rules.
- 30% higher client engagement: Advisors could deliver immediate, personalized recommendations in real-time.
Lessons for decarbonization tech leaders
Common pitfalls to avoid
- Relying on static dashboards for dynamic rebate logic
- Ignoring the complexity of multi-level rebate coverage
- Skipping hallucination safeguards in AI-driven recommendations
Advice for energy transition innovators
- Start with a PoC focused on speed to insight and accuracy
- Design for scale from day one with AWS-native architecture
- Build with real rebate and equipment data to ensure practical outcomes
Ready to accelerate your industrial decarbonization strategy with an AI copilot for decarbonization?
Let GoML help you transform fragmented energy data into instant, actionable guidance just like we did for Ampere Financial.
