At Kairos, we've been busy developing the platform! 

Here's what we've been doing:

Functionality

The most important thing - what we have delivered:

• Working software!
• Database capability to hold risks and related controls, user and organisation info, and posts. We'll be building out the risk model over the next few months.
• users can manage their “about” profile, adding an avatar, your role and a bit about themself - this will help the platform tailor its responses.
• Admin capability: manage organizations and users
• An "posts" interface to manage help content, announcements, and any other stories that need sharing - this page is actually simply a post that is regularly updated.
• The ability to view, add, update and delete risks and their associated information. The same for controls - and functionality to link/unlink them.
• The chatbot interface - this is the heart of the platform!
• And, most importantly … AI capability. The AI capabilities are expanding all the time:
• You can have a conversation with Kairos, your carefully trained "risk partner" assistant. 
• Kairos is focused on aspects of risk management and compliance, and will respond in that context.
• Responses will be relevant to your business and your role, as well as using your personal “about” to help make concepts relatable.
• Kairos also “knows” what the application can do, and is able to interact with it, for instance calling up a particular record for you based on your prompt.
• You can ask it questions about the risk and control data stored in the database. This is surprisingly powerful - have a look at the recent post on Retrieval Augmented Generation for more information.

Data

• A system for generating risks, controls, key indicators and other material for customers wanting to use a quick-start risk wizard.

Infrastructure and ways of working

Lots has gone on here …

• We now have a Singapore-based server to host the pre-production public-facing platform.
• A development environment at Kairos HQ with the latest equipment, funded by our backer!
• Implemented locally hosted virtual machines for user acceptance testing (UAT).
• Implemented source code management (git) to ensure an error-free application development
• We now have an industrial-strength Postgresql database back-end, moving away from the initial, lightweight sqlite database which served us well at the start of the development process (day-one startup). The code has also been tested with MS-SQL Server for portability).

Team development

The team has been busy upskilling! We've had to learn a lot …

• Python based web server development
• Bootstrap for clean web-styling. CSS, HTML and Javascript brush-ups! 
• Vector databases for doing data mining for advanced context - based search and retrieval
• using OpenAI's API to integrate AI into the platform
• A brush-up on git for source code management
• Some Linux and hypervisor refreshers

So what's coming?

The next phase will be to improve the capabilities of the AI assistant:

• Navigation: the assistant will take you directly to the pages and records that are most likely relevant to you, to update.
• Use of the local content. This will come in phases, as there is quite a lot involved:
• Phase 1 - Using the store of risks to suggest updating, vs new risks. This is a slightly contrived use-case as not all assistance requests will relate to risks.
• Subsequent phases - the assistant will figure out what the request relates to and use the appropriate data.
• Functional assistance. This is the final stage of the chatbot implementation:
• The assistant will help you actually update the system, advising you on where to go, taking them there, and helping you with context-relevant prompts to enter the best quality information.
• Monitoring and reporting: the final AI phase will involve using AI to generate meaningful dynamic information based on the up-to-date information in the system!
• More app functionality, including the quick-start risk wizard.

Options under consideration:

• Streaming the AI answers. This gives a more natural feel, as the user can start to read as soon as answers are generated. However, the current model calls for the answer to be post-processed to relevance and abuse, so this is not possible at present.
• Techie stuff: moving to micro services for easier development and debugging
• How to enhance the risk wizard experience through AI.

Technologies and standards used in this website:

Opensource - Thanks to so many people

SocketIO

Sentence Transformers

Flask

Jinja2

Bootstrap

Celery

PostgreSQL

Tagify

Bokeh

Milvus

Pickle

Redis

Python

ReportLab

Quill

SQLAlchemy

Standards - of course …

PDF

XML

YAML

CSS

HTML

Markdown

JSON

Javascript

Subscriptions

Linode

OpenAI

Github