Which types of deployments does gk support cloud or on-premise?
GK offers three different types of installations depending on your needs:
On-premise - installed in a firm's own data center, typically on a Virtual Machine.
Virtual Private Cloud (VPC) - for those firms leveraging AWS, a VPC is an attractive option, where you provide the network security and we handle software maintenance.
GK Public Cloud - for those customers desiring the most ease of use and zero maintenance burden, we offer our GK cloud on AWS and Google Cloud Platform (GCP).
What is a VPC?
A virtual private cloud (VPC) is an on-demand configurable pool of shared computing resources allocated within a public environment, where it provides a level of separation between different organizations leveraging compute resources.
What is Docker?
Docker is a container engine which uses the Linux Kernel features like namespaces and control groups to create containers on top of an operating system and automates application deployment on the container. GK on-premise solutions make heavy use of this technology. Find out more about Docker.
What is Kubernetes?
Kubernetes is an open-source container-orchestration system for automating application deployment, scaling, and management. GK on-premise solutions make heavy use of this technology. For our VPC customers, we leverage AWS EKS service. Find out more about Kubernetes
What are typical gk compute & memory requirements?
Linux Virtual Machine or Cloud Hosted Machine, 12 Logical CPU, 16 GB RAM, 200 GB disk.
Systems processing a large amount of chat / audio data would require resource requirements run through our resource calculators.
On what flavour of Linux can I install gk?
GK supports the following flavours of Linux: RedHat, Ubuntu, Debian and Centos.
Are all gk components Docker containers?
Yes, all are Docker containers
Which container systems does gk support?
GK supports EKS (AWS), GKE (GCP), AKS (Azure) and OpenShift (Red Hat)
Can we use Helm package manager to install software or is it installed manually?
GK provide both helm installs or raw Kubernetes manifests - whatever you prefer.
Which database management system does gk use? Are other databases supported, is it included part of a black box installation?
GK use PostgreSQL DBMS. We package it with the deployment, so you don't need to provide your own DB.
Does gk have specific storage requirements?
PostgreSQL and Redis containers each require a persistent volume.
Is Focus is a desktop application or browser based?
Focus Studio is an OpenFin desktop application. We have a browser-based Focus UI
What are Private Wires?
Private Wires provide always on connections between traders, brokers counter-parties and other market participants.
Does gk work out of the box?
For many customers, our out-of-the-box base model meets their needs. However, we have found that for a minimal investment in error corrections and training, you get a substantially better experience from customization.
What asset classes do we cover?
GK covers over 35 product classes within a broad range of asset classes, including Credit, Commodities, Rates, Equities, and FX.
Does gk understand multiple languages?
Our models are trained on over 200 Global English accents, as well as 35 financial product class languages that allow for product details to be extracted from non-english text messages. Our engine allows easy onboarding of new languages through our partnerships with data providers.
Can we extend your models?
Yes. We offer an SDK to help you extend our base models.
How will we maintain GK models once in production?
Firms can maintain both Chat & Audio models via GK's Focus Studio interface in production. You can go into Focus Studio, look at a sample set of your data and then tag and validate your data. It is from the corrections you are able to evaluate and score accuracy on your data set. All models are deployed at above human level accuracy.
Can gk help us customize and leverage those base models by integrating them into other systems and workflows?
Yes. We offer fee-based professional services in data science and development. Let us wire up our solution to your Enterprise.
Does the Discovery SDK have built-in entity support like countries, geographies, person, number, date?
Yes there are some built in categories, but for those that do not exist, our SDK allows you to create your own custom entities.
How do we know the quality of new model is actually improved ?
Our word error rate will decrease. The initial and final word error rate (WER) are listed in the our training systems' logs.
What is Automatic Speech Recognition (ASR) used for? I don't quite understand what is WER calculations
Automatic speech recognition just means transcription by a machine. Word error rate (WER) is the total number of word insertions, substitutions, and deletions compared to a truth transcript. It is a way to measure accuracy. Low word error rate means high accuracy. You can consider accuracy to be 100% - WER, so 20% WER is 80% accurate. Humans are 85% - 95% accurate at speech recognition, and therefore have 5% - 15% WER.
Does GK automate the transfer chat/audio files into Focus Studio?
Yes, chat and audio files can be fed directly via our API for automated processing.
Is there a data or character input limit to how much data can be ingested into GK?
There is no character or data limit - Chat / Audio files can be ingested ranging from a few MB to a few GB in size.
What types of text data does gk ingest?
We are data agnostic: we process chat, email, call notes, research articles and in a variety of formats.
What is a text classifier ?
A machine-learned classifier trained on thousands of examples provides one of the following labels: general, inquiry, quote, trade, market colour. Note that there isn't an exact "taxonomy" -- it's trained from data labelled by our network of SMEs.
How accurate are GK reports and can chat and audio be processed as one report?
Our models are trained to be at or above human level accuracy, and the combination of the accuracy of many models results in report accuracy around 90- 95 %. For example, a missed opportunities report would extract 90 - 95% of missed opportunities.
How does Bloomberg's chat NLP report integration work?
Most firms receive a dump of their Bloomberg chat/email message data at the end of the day in an .xml or .pst file format. Our API ingests this file directly (no preprocessing required), extracting meaningful metrics, product mentions, prices, missed opportunities plus other insights, into the reports. These reports are typically emailed by our system to select users within a firm or available as CSV or PDF downloads.
What is trade negotiation ?
Business logic rules are enforced to track the trade life-cycle. Note: "client" is defined as anyone with a firm name which does not match their own.
How does gk define an Inquiry?
It is a Client request about a financial product.
How does gk define a quote?
A firm's personnel mention of a product, price, or size. this will include axes / runs / etc.
How does gk define trade?
trade confirmation message by either party enforced to come after a quote. these are typically "i'll take it", "done", "etc"
How does gk define market color?
There are two categories of content here. (1) market analysis, insight, or other financial content (2) price updates (i.e. XYZ trading at 10.5)
How does gk define general/social?
All content outside of market_color, trade, Inquiry, quote.
What is an IOI ?
IOI is an Indication of Interest. In finance IOI demonstrates a buyer's non-binding interest in a security in the stock market
What is chat sessionization?
Tying an inquiry to subsequent quotes and a trade
What is chat thread disentaglement?
In a multiparty chat conversation, understand which speakers were discussing a particular topic