CASE STUDY

AI FOR FINANCIAL SERVICES

ecommerce

Stocks      Crypto      Fx Currencies

In capital markets, sentiment is everything. Imagine, as a trader, being able to deploy thousands of analysts to read every social post, news report and analyst reports, in real-time, before trading.   That’s what ElliSense’s AI does.

Detailed Description

Powered by artificial intelligence, this simple to use application gives an accurate reading of market sentiment that financial professionals can use to inform every trade. Market sentiment changes quickly and often unpredictably. The smartest trades, whether stocks, crypto, or forex, rely on up-to-the-minute sentiment analysis. Time is of the essence, which is why ElliSense gathers and digests many thousands of data points per second to give traders the intel they need to make smart decisions


The Artificial Intelligence (AI) analyses messages from financial analysts for various assets including currencies, crypto, and stocks. Using natural language processing (NLP) and sentiment analysis, the AI, with its 100+ million parameters, can classify analyst reports, news and social posts as bearish, neutral and bullish. The dataset uses over a quarter of a million messages that have been manually labeled to train the AI model. Currently it achieves over 97% accuracy and is getting better by the day.

After aggregating this information, ElliSense produces real-time indices of whether the market considers an asset to be bearish, neutral or bullish. While not a crystal ball, the results are surprisingly effective.

The ElliSense application is accessible from a laptop, tablet and mobile device or via an API framework for trading robots.

400+

Analysts

30 SECONDS

After a post is published, it is analyzed and integrated into our indices

25M+

Posts analysed per month

Classifies posts with over

97% ACCURACY

Dataset of more than a

QUARTER MILLION

Posts

Artificial Intelligence with

100+ MILLION

parameters

ecommerceecommerce

Main Features

User management (On boarding, Off Boarding , communication tools, etc )

CRM connectivity to Hubspot

ecommerce

Payments via Stripe

ecommerce

Connection to real time market feeds

ecommerce

Connection to multiple real-time social posts & data sources

Real time market news feeds

Desktop app, mobile app and partner widget interfaces

System Monitoring tool (uptime, capacity, optimisation)

Dedicated AI neural networks using NLP and Sentiment Analysis to analyze social posts

Use of different AI technics to optimize meta parameters

Producing real-time sentiment indices for 350+ financial assets

Alerts and notification management

API framework for developers of trading robots

Historical data visualization

CMS to manage FAQ, tutorials etc.

ecommerce

Software & Technical Architecture

The frontend is written in React JS, using the very latest technology such as browser notifications. This frontend exchanges information with the backend through REST API for the classic service requests, and with Websockets for the real-time information.

The backend is written in Java for the standard services. Each service is encapsulated in docker containers for scalability and portability.

We have used different AI techniques in building the engine for this product. All development was done in Python, using packages such as PyTorch and TensorFlow.

We have trained a Neural Network made of 125 Million parameters to analyze and interpret social media posts about financial assets. We were able to reach an unprecedented 97+% accuracy. This Neural Network uses the same technology as ChatGPT.

We have also used some other AI technology (regression, Montecarlo, topic modelling etc) for example to optimize the sentiment index hyper parameters or to categorize social posts.

Technical Stack

Frontend React JS + node JS

BackendDocker, Java +Junit for unit tests

AI Python (Pytorch & Tensorflow)

Database Mysql, Adminer

Communication REST Api & Websockets Traefik (Reverse DNS)

MonitoringnetData, custom toolling Cloud: AWS, GCP

UX & Product Design

The challenge with data is how to present it and how to explain it. Users need to trust an interface and we can build trust by having an intuitive, easy to use UI that fits into their workflows.

Like any great UX process, we start with discovery and understand what our users need. We let them tell us about the good and the bad of their current tools and watch and ask them about their daily tasks to understand the define the problem that needs solving. From there we come up with solutions and test test test, iterate, iterate, iterate until we give them the product solution that they didn’t know they needed.

With complex data applications we need to let them explore, but also need to give them quick bites of information that they can use to get their work done. This paradigm is know as the SNACK, BITE, MEAL approach.

Our UX Design Process

Discovery User / Stakeholder interviews, competitive research, workshops, heuristic evaluations, persona development

Define Define the problem(s) so we know what we need to solve for using user centered design thinking

Design Design solutions for the problem using wireframes & rapid prototyping tools

Test Test our solutions in low fidelity / workshops and monitor results to see if we are on the right track, if not return to define

Production Once we have our solution tested and are moving forward, we design in high fidelity for handoff to development

team

Consulting Engagement

In order to deliver ElliSense, Spring AI set up teams of experts in the disciplines of:

team

UX & Product
Design

Frontend
Development

Backend
Development

Cloud &
Security

Data
Science

Working closely with the client, our team was able to successfully deliver this ground breaking project. The first features were delivered within a few weeks. To ensure a successful project, we made investments in methodologies.

With design and project management we used well known methodologies, for AI we created and customised our own methodology from standard Agile.

THE FINAL PROJECT

hero