AI Concepts we can advise on and implement

• Security – paramount when using AI, ensuring that models do not have access to credentials and limit access to you local systems.
• Cost Management – that is managing how much text is back and forth to the AI model and effectively selecting suitable models for a given task.
• Tools – functions on your computer or system that the AI can invoke, for example reading a file or sending a slack message.
• Agents – AI models combined with tools, operating in loop, where by the model responses to user requests and decides on the tools to call on, it will consider the results of the tools and may ask for further user input, call another tool or produce a final result.
• Harness – we can provide a harness that wraps your AI capabilities with credential management, limiting what data, tools and models your agent or people have access to.
• API – this is the access layer that you pay OpenAI, Google or Anthropic for so you can sends messages and requests to the AI and receive it’s results.
• Vector database – A database that represents your text content as numbers, this allows for effective retrieve of your data with the correct semantic meaning, i.e. it understands the meaning of questions so that appropriate data can be extracted from your datasets and passed to the AI model for its consideration.
• Evals – this is the term for testing AI products, that is how to measure it is doing what you require it to do, for example can it effectively parse emails into the correct work flow. Effective evals are important when testing alternative models or varying effort levels in an effort to reduce costs.
• Scheduling – sounds simple but it is quite often a challenge to ensure that event driven actions are handled correctly, there are a number of patterns and services that can assist to ensure that sequential or concurrent pipelines are running correctly whilst managing failures and duplicates.

TECHNOLOGY

Details for a technical audience

Tech stack outline

Node, Javascript and React

The classic framework for rapid application development, a full ecosystem exists for testing, authentication, reactive UI, database access and easy integration with third parties services

C#, ASP.NET and SQL Server

A technology suite familar to some of our enterprise clients, although we see modern Microsoft frameworks, languages and service, particular MVC and WEB API gaining popularity in all types of organisations. A framework that facilitates quick, robust development whilst promoting modern practices such as dependency injection and domain driven development this is now our preferred stack for modern web development.

Python, Postgress Db and AWS

We have consistenly seen a rise in requests for data centric applications, this is where Python is a natural choice, with its mature library of packages for surfacing, presenting, analysig and exploring data.

AI Models

Currently we’re focussed on offering AI integration through the OpenAI API, we are very familar with the API, services and model behaviour. Behind the scenes though we continually review and experiement with other frontier models to compare their offering in terms of capability and value. We’re also assessing some locale models for particular use cases such as PDF or image parsing for use cases such as invoice processing and document auditing.

Technology

The landscape of services and technologies associated with modern software development and AI deployment is broad and evolving, here we’ve described the high level, but we’d be more than happy to discuss any other aspects you maybe aware of or actively considering, perhaps graph databases, duckdb, decisions API, event bridges, cloud file buckets, telemetry etc etc, it’s part of our function to keep up and assess applicability.