Empowering Technicians Throughout Each Retailer
7‑Eleven’s upkeep technicians maintain shops working easily by servicing a variety of kit — from meals service home equipment and refrigeration items to gas dispensers and Slurpee machines. Every restore depends on the technician’s data and rapid entry to supporting paperwork, akin to service manuals, wiring diagrams, and annotated photos.
Making a Unified and Sooner Method for Technicians to Discover Tools Data
Over time, tools documentation has developed to incorporate a number of codecs, unfold throughout numerous places. This makes it tougher for Technicians to find the knowledge they want rapidly. Furthermore, when encountering unfamiliar tools, elements, and many others., Technicians would typically depend on chat or e-mail to get help from their friends.
As such, a possibility to streamline how info is accessed, shared, and many others. was recognized; in the end leading to extra constant help for retailer operations.
Constructing the Technician’s Upkeep Assistant (TMA)
To deal with these challenges, 7‑Eleven envisioned an AI‑powered assistant that might:
- Retrieve exact solutions from upkeep paperwork.
- Determine tools elements from photos and recommend associated supplies.
- Combine seamlessly inside Microsoft Groups.
Partnering with Databricks, 7-Eleven developed the Technician’s Upkeep Assistant (TMA), an clever answer that integrates doc retrieval, imaginative and prescient fashions, and collaboration right into a streamlined workflow.
Doc Storage and Indexing
All related upkeep paperwork had been uploaded to a Unity Catalog Quantity, which manages permissions for non-tabular knowledge, akin to textual content and pictures, throughout cloud storage.
Utilizing Databricks Vector Search, the event group applied Delta Sync with Embeddings Compute. They generated vector embeddings utilizing the BAAI bge-large-en-v1.5 mannequin, and served them by means of a Vector Search endpoint for high-speed, low-latency retrieval.
Microsoft Groups Integration
Technicians entry TMA straight by means of Microsoft Groups. A Groups Bot routes every question by means of an API layer that orchestrates calls to Databricks Mannequin Serving. The assistant supplies contextual solutions, matches documentation hyperlinks, and suggests related elements straight within the chat window.
Routing Agent and Sub‑Agent Design
A Routing Agent determines whether or not a technician’s question is document-based or image-based, directing it to the right sub-agent:
- Doc Query and Reply Agent
- Technicians can use pure language queries inside Groups. With Claude 3.7 Sonnet by way of Databricks Mannequin Serving, the system converts these queries into vector embeddings, searches the index, and returns context-aware solutions utilizing Retrieval-Augmented Technology (RAG). Technicians obtain responses immediately, even from lengthy manuals or tools guides.
- Picture Identification Agent
- Early variations used simple textual content extraction by way of Claude 3.7 Sonnet however yielded uneven outcomes. Engineers enhanced efficiency by tailoring prompts to technician workflows — overlaying product numbers, producer particulars, specs, security warnings, and certification dates.
- The extracted knowledge maps on to Delta Desk fields, linking visible references to the right paperwork within the vector index. This refinement produced extra correct and dependable half recognition.
Logging and Analytics
To take care of transparency and knowledge governance, all interactions — routing, queries, and picture requests — are logged in Amazon DynamoDB. A day by day Databricks Job extracts these logs, shops them in Delta tables, and powers a devoted AI/BI Dashboard.
The dashboard provides 7‑Eleven visibility into:
- Every day/Weekly/Month-to-month (see beneath) question quantity by technician.
- Most often looked for or serviced tools.
- Chatbot decision traits and latency.
- Correlation between TMA adoption and improved first‑time‑repair charges.
Migration from AWS to Databricks
The primary proof of idea utilized AWS parts, together with SageMaker, FAISS, and Bedrock, to host giant language fashions akin to Claude 3.7 Sonnet and Llama 3.1 405B. Whereas useful, this setup required guide reindexing, a number of indifferent companies, and launched latency.
To simplify its infrastructure, 7-Eleven migrated to a completely Databricks Agent Bricks answer, end-to-end, which resulted in accelerated response instances.
Key enhancements:
- Automated vector indexing with Databricks Vector Search.
- Unified knowledge governance and compute administration.
- Decrease latency and simplified observability by means of a single lakehouse structure.
Delivering Operational Impression
“From what I’ve skilled to this point, the Technician’s Upkeep Assistant has the potential to enormously enhance the pace, accuracy, and consistency with which our technicians entry vital documentation for preventive upkeep and tools restore,” mentioned James David Coterel, Company Upkeep Coach at 7‑Eleven.
By streamlining doc retrieval and lowering dependency on peer help, the TMA enhances technician confidence, improves first-time-fix charges, and cuts search time from minutes and even hours to seconds; straight lowering downtime and accelerating retailer readiness.
In parallel, shifting retrieval, embeddings, and inference from AWS to Databricks eradicated FAISS upkeep and EC2 load, reducing infrastructure overhead and bettering latency, which compounded into measurable operational financial savings and a extra constant buyer expertise.
Whereas the precise greenback impression remains to be being measured, the mixture of sooner first-time decision, fewer guide escalations, and decrease infrastructure overhead creates clear price avoidance on labor hours and unplanned tools downtime, each of which correlate strongly with retailer income safety and buyer expertise stability.
Future Enhancements
7‑Eleven plans to broaden TMA’s capabilities by means of:
- Video-based upkeep guides for visible and arms‑on studying.
- Multilingual help for international upkeep groups.
- Information‑pushed suggestions loops to repeatedly refine response accuracy and relevance.
Uncover how Databricks permits enterprises like 7-Eleven to construct clever assistants that combine knowledge, paperwork, and imaginative and prescient fashions on a single platform.
