Adding AI to your Application Network
At QuadCorps we are augmenting (no pun intended) our existing products and services with a subset of Artificial Intelligence (AI) called Retrieval Augmented Generation (RAG).
RAG applications provide business with new capabilities, insights and ways to interact with your existing data, using natural language to realise the benefits of AI.
RAG solutions require a strong API and automation foundation to be successful. QuadCorps is leveraging its extensive experience in API and Integration ecosystems and its expertise in business process automation to offer RAG based solutions and data pipelines.
What is Retrieval Augmented Generation
Think of RAG as applying the power of AI (specifically Large Language Models) to your own proprietary data, safely, without the need to train your own AI model (which is time consuming and expensive).
RAG is a way of processing large volumes of unstructured data (and structured data), and converting it into mathematical representations of that data (using AI). These mathematical representations can infer the meaning (semantics) of the data based on its context and internal relationships. This is very powerful.
Large Language Models are very good at processing the mathematical representation of the data, called Vectors, quickly and cheaply, providing new classes of capability.
Use-Cases and Business Benefits
AI is currently surrounded by a mass of hype. RAG is one of the use-cases that has substance and real world applications and benefits.
RAG solutions are capable of processing massive amounts of data much faster and more accurately than humans can.
RAG helps you gain new insights into your data. This is partly due to the fact that you can interact with your data using natural language making it more accessible to a wider non-technical audience.
RAG unlocks the value of your unstructured data (documents, PDF’s, websites, logs, images, video, audio etc.) You can feed old stored data into RAG solutions to gain new insights.
Future articles will focus on some of the specific use-cases for RAG.
RAG Vs Pre-Trained Models
Many businesses have been experimenting with AI models over the last 18 months. Models such as ChatGPT have almost become household names.
AI has been in the press for both its abilities and its drawbacks. RAG. solutions focus on the benefits of AI models and reduce substantially the risks and drawbacks of these models.
Pre-trained models like ChatGPT require massive amounts of training data. Training data comes from the internet and other sources. Training is both slow and expensive and most importantly doesn’t include your proprietary data.
Below is a table that shows the benefits of using your own data in RAG applications rather than relying on pre-trained models.
| RAG Applications | Pre-Trained Models |
|---|---|
| Use your existing proprietary data | Require large volumes of external training data |
| No need to train models | Training models is very slow |
| No need to train models | Requires large volumes of training data |
| No need to train models | Training models is very expensive |
| You always have access to your latest data. A RAG data pipeline can keep your AI data updated in near real-time. | Pre-trained models don’t know anything beyond the date their training ends. They can be updated (tuned) but this is both slow and expensive. |
| Hallucinations are greatly reduced as (your) data is always available to the model. | Prone to Hallucinate if they have no relevant data. This makes results less reliable. |
| Query response times relatively fast | Query response times are slow |
| Accuracy of results is high for queries about your data. | Accuracy of results depends on training data |
| Private: based on private LLM instances | Data leaks possible in public LLM instances |
AI is good at specific tasks and also has some drawbacks. RAG solutions are a way to leverage the benefit and minimise the drawbacks:
- Use your existing proprietary data no matter the format, structure or location
- Process large volumes of unstructured data much faster and more accurately than humans or traditional solutions.
- Semantic solutions are possible using Vector based data.
- Provide more accurate and relevant results than pre-trained models and reduce hallucinations.
- Return query results faster than queries against pre-trained models
- Are private and safe as they would use private instances of AI models which won’t leak your proprietary data to other users.
AI applications based on RAG should become just another building block in your composable architecture which powers agile, automated business processes.
RAG is a rich and deep subset of AI. Look out for additional articles that describe the benefits and use-cases of RAG in more detail.
QuadCorps Capabilities: Why RAG?
AI is all about data. RAG is all about your data.
Exposing, consuming, cleaning, transforming, moving and querying data is all a necessary part of a RAG data pipeline. As such Integration and API’s form the foundation of AI / RAG solutions.
A solid A.P.I. strategy greatly increases the odds of successfully implementing any AI or RAG. project.
There is no AI without API
QuadCorps are experts in designing architectures and building solutions that expose your business data and capabilities in a modular, composable manner.
Using Integration and API patterns, practices and technologies to obtain and process data into new automated processes, products and services, quickly, is what we do.
RAG is a natural extension of QuadCorps capabilities. We understand data, we understand API’s and Integration. We know how to automate processes. We already have the skills and experience required to deliver real world, automated, near real time RAG data pipelines.
We have spent the last few years learning the AI aspects of RAG (Retrieval & Generation - The ‘R’ and ‘G’ in RAG)
QuadCorps now possesses the capability to develop both RAG applications and to build specialist RAG data pipelines that automate the AI processing of data in real time.
(Future articles will delve into the details of whats involved in processing data for RAG applications)
QuadCorps AI Credentials
While this post is the first public announcement that we are complementing our existing services with a RAG capability, we have been busy in the background, identifying real word AI use-cases and delivering solutions to our clients.
We are currently building an AI platform for a large insurer and are looking at working with a bank to identify potential AI use cases.
Our AI solutions thus far include:
- Building a standardised data ingestion framework to support RAG functionality
- Building a LLM agnostic capability that allows customers to switch between AWS and Azure AI services
- Building a set of reusable capabilities for the integration of LLM’s into business applications, while enabling:
- Centralised management
- Governance
- Observability
- Enforcement of the guardrails that AI applications require
We also have use-cases around Fraud Reduction using C.C.T.V footage and Auto-Generation of Tag suggestions for an online news site.
We are developed partnerships with some of the best of breed AI vendors including vector database, API automation and Data Cloud platforms. We are investing in QuadCorps ability to deliver AI powered semantic solutions.
Is AI On Your Mind? Get in Touch…
If you have been wondering how your business can leverage AI and would like to talk it through with someone knowledgeable give us a call.
We are happy to have a chat and have one of our AI experts help you explore the benefits AI can bring to your business. No charge, no obligation, no pressure.
If you would like to get in touch please feel free to please contact me via linked-In (www.linkedin.com/in/vaughnhaybittle) or directly via e-mail at vaughn@quadcorps.co.uk
