01 4–8 weeks
RAG on Azure AI Search
Ingestion, hybrid index, Azure OpenAI with source citation and an evaluation set, deployed in your subscription.
AI inside your own Azure subscription
If your company already runs on Microsoft, the models, the indexes and the data can live in your own subscription. We build agents, RAG and LLM applications on Azure OpenAI and Azure AI Search, with your credentials and the usage billed to you.
3 years running Azure OpenAI and Azure AI Search in production · 2M+ users served by that system across ten countries
Built on
An engineering team that has already been on the other side.
Ninety seconds: who we are, how we work and what you get at the end.
From your company’s documents to a system that answers and acts
Contracts, price lists, email, the CRM: what you already have becomes sourced answers and agents that do the work. Pick a case and follow the path.
Example: a customer asks what their policy covers. The system answers with the exact page of the contract and opens the case.
01Your documents
Policy_wording_2026.pdf84 pp
Motor_exclusions.pdf12 pp
Claims_email1,240
ERP · policies38,400
Price_list_2026.xlsx118 rows
CRM · opportunities2,310
Proposals_202556 PDF
Sales_email4,800
Supplier_invoices3,900 PDF
ERP · purchase_orders4,110
Framework_contracts.pdf22 docs
Payments_email2,100
Connected by API or uploaded. Nothing leaves your infrastructure when the case demands it.
02Preparation
Normalise
Chunk
Tagproduct = motor_premium
Make searchable
84 pp → 612 chunks
03Search
94 of 100finds the right chunk
142 mstakes
Searches by meaning and by exact word, and filters by the tag before choosing.
04Sourced answer
Does it cover theft from a public car park?
Yes, with a 10 % excess. Source: 2026 policy wording, clause 4.3.
answers with source · 91 of 100
05Agent that acts
- Queried · policy 88213
- Read · clause 4.3
- Decided · open a claim
- Asked for approval · human
- Executed · created the case in the ERP
- Queried · CRM
- Queried · price list
- Drafted · proposal
- Asked for approval · manager
- Executed · sent and logged
- Read · invoice 9902
- Cross-checked · PO-4471
- Detected · discrepancy
- Escalated · human
- Logged · case opened
Swipe to follow the flow →
Why Azure
If you already run on Microsoft, the AI can stay there
The first question the security team asks is not which model you will use but where the data goes. When the company already has Microsoft 365, Entra ID and a Microsoft contract, Azure lets the model, the index and the logs sit under the same identity, the same policies and the same invoice.
It is not a better model: it is the same OpenAI models, served by Microsoft under the contract the company already signed. What changes is that the conversation with security and procurement has already happened, and that usually saves more weeks than any technical decision.
The pieces, inside your subscription
01
RAG on Azure AI Search
A hybrid index —by meaning and by exact words— with semantic reranking and metadata filters, measured with recall@k before it reaches production.
02
Azure OpenAI models with per-task routing
An inexpensive model classifies and summarizes, a capable one writes the final answer. Quota and cost per query stay visible.
03
Agents that query your systems
Deployed in your subscription, with permissions scoped to what each tool needs and human approval on steps with consequences.
04
Evaluation on every change
A set of real questions with their correct answers runs before every deployment, so you know whether an improvement actually improved anything.
01
Formats
Scope and timelines
02 2–4 weeks
Moving a pilot into your subscription
A prototype built on the public OpenAI API, moved to Azure OpenAI and Azure AI Search with quality measured before and after.
03 1 week
Diagnostic of what you already run on Azure
Retrieval and answers measured separately, cost per query, and failures ranked by impact with their fix.
When not to hire us
If you are not on Azure, we will not move you there for this. The same architecture runs on AWS or on the provider’s direct API, and a cloud migration is not justified by an assistant.
If what you want is for the team to draft emails and summarize meetings, Microsoft 365 Copilot already does it and nothing needs to be built. And if the work is the infrastructure itself —networking, landing zone, server migration— we are not the team: we build what runs on top of it.
Related services
RAG systems
RAG that answers well — and knows when to stay quiet
Chunking, reranking and hybrid search, evaluated with recall@k and NDCG.
AI agents
Agents that do the work, not just talk about it
LangGraph orchestration, durable state and human approval on steps with consequences.
Project rescue
Your AI answers badly. It is almost never the model’s fault.
The system is already in production and answers badly. We measure it and determine what to fix.
AI on AWS
AI inside your own AWS account
Amazon Bedrock and Claude inside your AWS account.
Key takeaways
Four things before the call
- 01
Microsoft Foundry —called Azure AI Foundry until November 2025— is where Microsoft now brings together the Azure OpenAI models, the agent service and evaluation, all inside the company’s own Azure subscription.
- 02
With Azure OpenAI, what is sent to the model is not used to train it and is not shared with OpenAI, and the Azure AI Search index lives in the same subscription as the rest of the data.
- 03
Quarl has spent three years running Azure OpenAI and Azure AI Search in production, in a RAG system serving two million users across ten countries.
- 04
Resources, keys and usage stay in the company’s name, billed directly by Microsoft, with no Quarl account in the middle.
This is where the data comes in
An AI system is worth what its sources are worth. These are the standard connectors; anything with an API or a database connects the same way, and what has no API is handled by file.
SAP
Enterprise ERP
Oracle
ERP and database
NetSuite
Cloud ERP
Salesforce
CRM and service
HubSpot
CRM and marketing
PostgreSQL
Database and pgvector
Microsoft SQL
Database
Snowflake
Data warehouse
BigQuery
Google data warehouse
Databricks
Data platform
Redshift
AWS data warehouse
Synapse
Azure data warehouse
Supabase
Managed Postgres
Workday
Payroll and HR
QuickBooks
Accounting
Sage
Accounting and ERP
Xero
Cloud accounting
Shopify
Catalogue and orders
WooCommerce
Catalogue and orders
Magento
Catalogue and orders
Stripe
Payments and subscriptions
Google Drive
Documents and folders
CSV y Excel
Flat files
Nothing in this category
A RAG system in production: three years, two million users
We built and operated the assistant for a loyalty platform serving more than two million active users across ten countries.
We built the full pipeline: document ingestion and normalization, chunking, embedding generation, vector store on Azure AI Search and Pinecone, and retrieval with grounded generation on LangChain.
We held it above 99% availability for three years.
RAG systems →Frequently asked questions
01 What is Microsoft Foundry, and how does it relate to Azure AI Foundry?
It is the same product under a new name: Microsoft renamed Azure AI Foundry to Microsoft Foundry in November 2025. It brings together the models —including Azure OpenAI—, the agent service, evaluation and content filters in one place, inside the company’s Azure subscription. If a proposal mentions either name, it is talking about the same thing.
02 Is our data used to train models?
No. Azure OpenAI is a Microsoft service: what is sent to the model is not used to train it and is not shared with OpenAI. The search index, the documents and the logs stay in resources in your subscription, under the Entra ID permissions you already manage. We also filter personal data out before it reaches the model when the use case does not need it.
03 Do you use the Foundry agent service or LangGraph?
It depends on what the agent does. If the workflow fits what the Foundry agent service provides —conversations with history, tools and knowledge from Azure AI Search—, it is worth using: it is less code to maintain. If it needs a graph of steps with fine-grained control and human approval in the middle, LangGraph deployed in your subscription gives you that control. Both run inside Azure, and the choice is justified in the proposal.
04 Who pays for the Azure usage?
You do, directly to Microsoft, under your subscription or your enterprise agreement. Quarl does not resell usage or put an account of its own in the middle. What we do deliver is the measured cost per query, so the Azure invoice is not a surprise when usage grows.
05 Can you work on what we already have on Azure?
Yes, and it is the most common case: an Azure AI Search index or an Azure OpenAI deployment that already exists and answers badly. We start by measuring it with real questions and their correct answers, as in a rescue, and the fix is made on your resources, without rebuilding what works.
Fifteen minutes. A concrete answer.
It gets fixed, it gets built, or it is not worth it. And if the diagnostic does not reach three actionable findings, it is not charged.