Empowering technicians in every store
7‑Eleven’s maintenance technicians keep the store running smoothly by servicing a wide range of equipment, from food service equipment and refrigeration units to fuel dispensers and Slurpee machines. Every repair depends on the technician’s knowledge and immediate access to supporting documents such as service manuals, wiring diagrams and annotated images.
Creating an integrated and faster way for technicians to obtain equipment information
Over time, tool documentation has evolved to include multiple formats spanning different locations. This makes it difficult for technicians to quickly find the information they need. Additionally, when faced with unfamiliar equipment, parts, etc., technicians often rely on chat or email to get support from their peers.
Thus, an opportunity was identified to streamline information accessing, sharing, etc.; Ultimately there was more consistent support for store operations.
Creation of Technician Maintenance Assistant (TMA)
To address these challenges, 7‑Eleven envisioned an AI-powered assistant that:
- Get accurate answers from maintenance documents.
- Recognize equipment parts from images and suggest related content.
- Integrate seamlessly within Microsoft Teams.
Partnering with Databricks, 7-Eleven developed Technician Maintenance Assistant (TMA), an intelligent solution that integrates document retrieval, vision models, and collaboration into one streamlined workflow.
Document Storage and Indexing
All relevant maintenance documents were uploaded to a Unity Catalog VolumeWhich manages permissions for non-tabular data such as text and images in cloud storage.
Using Databricks vector searchThe development team implemented delta sync with embedding compute. He devised vector embeddings using BAAI bge-large-en-v1.5 modeland served them through a vector search endpoint for high-speed, low-latency retrieval.
Microsoft Teams integration
Technicians access TMA directly through Microsoft Teams. A Teams bot routes each query through an API layer that orchestrates calls to Databricks model servingThe assistant provides relevant answers, matches documentation links, and suggests relevant parts directly in the chat window,
Routing Agent and Sub-Agent Design
A routing agent determines whether a technician’s query is document-based or image-based, directing it to the correct sub-agent:
- Document Questions and Answers Agent
- Technicians can use natural language queries within teams. With Cloud 3.7 Sonnet via Databricks Model Serving, the system converts these queries into vector embeddings, searches the index, and returns context-aware answers using Retrieval-Augmented Generation (RAG). Technicians get instant responses, even from lengthy manuals or equipment guides.
- image recognition agent
- Early versions used direct text extraction via Cloud 3.7 Sonnet, but yielded uneven results. Engineers enhance performance by creating prompts for technician workflow – including product numbers, manufacturer descriptions, specifications, safety warnings and certification dates.
- The extracted data maps directly to delta table fields, connecting view references to the correct documents in the vector index. This refinement produced more accurate and reliable part identification.
Logging and Analytics
To maintain transparency and data governance, all interactions – routing, queries and image requests – are logged Amazon DynamoDBA daily Databricks job extracts these logs, stores them in delta tables, and powers a dedicated AI/BI Dashboard,
The dashboard gives 7‑Eleven visibility into:
- Daily/Weekly/Monthly (see below) query volume by technician.
- Most frequently searched or serviced devices.
- Chatbot resolution trends and latency.
- Correlation between TMA adoption and improvements in first-time rates.
Transfer from AWS to Databricks
The first proof of concept used AWS components including SageMaker, FAISS, and Bedrock to host large language models such as Cloud 3.7 Sonnet and Llama 3.1 405b. While functional, this setup requires manual reindexing, many separate services, and latency.
To simplify its infrastructure, 7-Eleven fully migrated to the Databricks Agent BRICS solution from end to end, resulting in faster response times.
Major Improvements:
- Automated vector indexing with Databricks vector search.
- Integrated data administration and calculation management.
- Low latency and simplified observability through a single Lakehouse architecture.
provide operational impact
“From what I have experienced so far, Technician Maintenance Assistant has the potential to significantly improve the speed, accuracy and consistency with which our technicians access critical documentation for preventive maintenance and equipment repairs,” said James David Cottrell, corporate maintenance trainer at 7-Eleven.
By streamlining document retrieval and reducing reliance on peer support, TMA boosts technicians’ confidence, improves first-time-fix rates, and reduces search times from minutes or even hours to seconds; Directly reducing downtime and speeding up store preparation.
In parallel, moving retrieval, embedding and inference from AWS to Databricks eliminated FAISS maintenance and EC2 load, reduced infrastructure overhead and improved latency, leading to measurable operational savings and a more consistent customer experience.
While the exact dollar impact is still being measured, the combination of faster first-time resolution, less manual escalation, and lower infrastructure overhead avoids obvious costs on labor hours and unplanned equipment downtime, both of which correlate strongly with store revenue security and customer experience consistency.
future enhancements
7‑Eleven plans to expand TMA’s capabilities:
- Video-based maintenance guides for visual and hands-on learning.
- Multilingual support for global maintenance teams.
- Data-driven feedback loops to continuously refine response accuracy and relevance.
Find out how Databricks enables enterprises like 7-Eleven to create intelligent assistants that integrate data, documents, and vision models on a single platform.
Explore Databricks AI Solutions
