AtWork Group
Knoxville, TN
Staffing and Workforce Solutions
Amazon Bedrock (Titan Embed v2), Amazon Bedrock (Amazon Nova Pro), Amazon OpenSearch Service, Amazon ECS Fargate, Amazon Textract, Amazon EventBridge, AWS Lambda, Amazon API Gateway, Amazon Dynamo, Amazon S3, DB, AWS Systems Manager Parameter Store, Amazon CloudWatch, Python, FastAPI, Mangum, SharePoint (Microsoft)
AtWork Group
Knoxville, TN
Staffing and Workforce Solutions
Amazon Bedrock (Titan Embed v2), Amazon Bedrock (Amazon Nova Pro), Amazon OpenSearch Service, Amazon ECS Fargate, Amazon Textract, Amazon EventBridge, AWS Lambda, Amazon API Gateway, Amazon Dynamo, Amazon S3, DB, AWS Systems Manager Parameter Store, Amazon CloudWatch, Python, FastAPI, Mangum, SharePoint (Microsoft)
AtWork Group is a national staffing and workforce solutions company operating through a franchise network across the United States. Franchisees lacked a fast, reliable way to access the company’s extensive operational documentation stored in SharePoint, leading to wasted time, overburdened support teams, and inconsistent policy interpretation across locations. Avahi designed and deployed a serverless Retrieval-Augmented Generation (RAG) smart assistant on AWS that gives every franchisee instant, accurate answers to operational and policy questions in plain English. The result is a self-maintaining knowledge assistant that continuously syncs with AtWork’s SharePoint content, eliminating manual overhead and delivering consistent, grounded responses at scale.
AtWork Group is a nationally recognized staffing and workforce solutions company headquartered in Knoxville, Tennessee. Operating through a distributed franchise model, AtWork partners with businesses across the United States to provide temporary, temp-to-hire, and direct placement staffing services. The franchise network model means that individual owners and their teams are responsible for running day-to-day operations in alignment with AtWork’s corporate policies, IT standards, and onboarding procedures, making fast, accurate access to operational guidance a critical business function.
AtWork’s policies, IT standards, onboarding procedures, and operational guidelines lived in SharePoint document libraries, a system wellsuited to document storage, but poorly suited
to on-demand information retrieval. When franchisees had questions, they were left to manually search through SharePoint without always knowing which library or document contained the answer. Many escalated to support teams who were repeatedly fielding the same questions, pulling internal resources away from higher-value work.
For newly onboarded franchisees, the challenge was even more acute. The volume of documentation was overwhelming, and there was no efficient way to navigate it quickly during a busy workday. This created compounding operational problems: franchisees spent time searching for answers instead of running their businesses, support teams were bogged down with repetitive, self-serve questions, and inconsistent interpretation of policies across franchise locations was introducing operational variation that AtWork needed to eliminate.
Left unaddressed, these problems would only intensify as the franchise network grew. More franchisees mean more questions, more support burden, and greater risk of policy inconsistencу,
all of which erode the operational standards and brand consistency that a franchise model depends on.
AWS provided the ideal foundation for this solution because of its tightly integrated suite of Al, serverless compute, and managed data services. Amazon Bedrock offered access to best-inclass embedding and generative Al models, including Titan Embed v2 and Amazon Nova Pro, without the overhead of managing Al infrastructure. Amazon OpenSearch Service provided a
production-ready vector database capable of semantic similarity search at scale. Together with ECS Fargate, Lambda, and EventBridge, AWS enabled a fully serverless architecture that
requires no infrastructure management, scales automatically with demand, and keeps operational costs aligned with actual usage.
AWS also made it straightforward to build the incremental ingestion pipeline that is central to keeping the knowledge base current. DynamoDB’s flexible data model was a natural fit for
tracking ingestion state and storing conversation history, while Amazon Textract handled the document extraction challenges inherent in real-world corporate SharePoint libraries – PDFs,
scanned files, and images included.
Avahi is a premier-tier AWS Partner with deep expertise in designing production-grade Al and data solutions on AWS. AtWork needed a partner who could move from concept to a fully deployed, operational system, not just an experiment, and who understood how to architect for the specific constraints of a franchise operational model: distributed users, high query volume, and a knowledge base that changes continuously as policies evolve.
Avahi brought the right combination of AWS service depth and applied Al architecture experience to design a solution that was not only technically sound but operationally complete. From the incremental ingestion pipeline to the multi-turn conversation memory to the Microsoft Teams integration path, every component was built with AtWork’s real-world franchise use case in mind, not adapted from a generic template.
Avahi built a fully serverless RAG smart assistant on AWS that allows AtWork franchisees to ask natural language questions about company policies and operational procedures and receive
accurate, context-aware answers in seconds. The system operates through two parallel, automated workflows: an ingestion pipeline that keeps the knowledge base current, and a query
pipeline that handles real-time franchisee questions.
Ingestion Pipeline: Amazon EventBridge triggers a scheduled ECS Fargate task that connects to AtWork’s SharePoint library, detects newly added or modified documents, and downloads them automatically. Raw documents are stored in Amazon S3, then passed through Amazon Textract for intelligent text extraction — handling PDFs, scanned documents, and images accurately. Extracted text is split into optimized chunks, converted into semantic vector embeddings using Amazon Bedrock (Titan Embed v2), and indexed in Amazon OpenSearch Service. A DynamoDB timestamp record ensures that only new or changed documents are processed on each run, keeping the knowledge base continuously current without any manual intervention.
Query Pipeline: When a franchisee submits a question, Amazon API Gateway receives the request and routes it to an AWS Lambda function running inside a VPC. Lambda embeds the query using Bedrock, performs a vector similarity search in OpenSearch to retrieve the most semantically relevant document chunks, and pulls the last 10 conversation turns from DynamoDB to maintain multi-turn context continuity. The retrieved content and conversation history are passed to Amazon Nova Pro via Bedrock, which generates a structured, natural language response. The answer is returned to the user through API Gateway, and the conversation is persisted in DynamoDB for continuity in follow-up questions.
The architecture is fully serverless – there are no servers to provision or maintain. AWS Systems Manager Parameter Store handles secure centralized configuration management, and Amazon CloudWatch provides logging, monitoring, and performance metrics across both pipelines.
The incremental ingestion design is a particular differentiator. Rather than reprocessing AtWork’s entire document library on every scheduled run, the pipeline fetches only what has changed. The initial ingestion processed 150 documents in approximately one hour. Every subsequent run processes only incremental updates, dramatically reducing compute cost and ensuring the knowledge base is always synchronized with the latest SharePoint content without human involvement.
AtWork franchisees now have instant access to accurate operational guidance without manually searching SharePoint or escalating to support teams. The assistant grounds every response in
AtWork’s actual documentation, eliminating the risk of inconsistent policy interpretation across franchise locations. Support teams are freed from repetitive, self-serve questions, and newly
onboarded franchisees have a fast, accessible way to navigate a large documentation library from day one.
The solution is also self-maintaining.The knowledge base updates automatically on a schedule, requiring zero manual intervention when AtWork’s SharePoint content changes. This ensures that every answer the assistant delivers reflects AtWork’s current policies, not a static snapshot.
Let’s explore your high-impact AI opportunities together in a complimentary session
Let’s explore your high-impact AI opportunities together in a complimentary session
Let’s explore your high-impact AI opportunities together in a complimentary half-day session
