Introducing CansoAI Platform

Platform to accelerate ML Systems development & decisioning

We are excited to introduce Canso, a platform to help ML teams accelerate feature authoring, orchestration, deployment and monitoring, through a no-low code approach. This is our first step (MVP) towards a larger vision of simplifying ML decisioning and development.

Canso Platform

ML teams, working on complex systems like fraud detection, recommendation systems are often found either struggling with engineering tasks such as feature engineering, orchestration, achieving sub-second latency, monitoring alerts etc. or dependent on engineering function for the same. In between all this, more important initiatives around model performance improvement that bring the most value takes a back seat.

In our experience of working with several companies ranging from early stage startups to large enterprises, we noticed a common problem that these tasks take up more than 60–70% of the time and cost that goes into any ML project. This significantly reduces the speed of experimentation and delays value creation. For e.g. It can easily take a whole quarter for a small ML team working on real-time fraud detection to upgrade the model with a feature based on a new hypothesis. There are several feature platforms available but due to piecemeal nature of the product, high cost and not having quality support, numerous smaller companies are unable to get the right value.

Canso solves this problem by providing a seamless design abstracting away complex data engineering workflows, a team of experts to help you with ML systems whenever required and making all this available at an affordable cost. Some of the features available in MVP:

Canso Feature Platform

Define & Deploy Features in Minutes -

1. Declarative framework to help Data scientists author features using a no-code approach..
2. No need to write Airflow DAGs; Canso supports managed orchestration.
3. One-click dry run & deployment

Integrations with your existing observability suite -

1. Canso comes integrated with Grafana, Prometheus, Loki and Slack
2. Built-in visualizations of health KPIs for your pipelines and services
3. Automatic slack alerts for the owners of features and services

Feature Discovery & Management -

1. A Feature registry to help teams discover and use features easily
2. Easily create derived features by using existing ones saving hours of effort

With Canso, Data Scientists & Engineers can easily get features/rules in production in minutes which otherwise is weeks of effort. This not only reduces ML cost by 40–50%, but also unlocks accelerated experimentation. Teams can conduct 10–12 experiments in a year, compared to 3–4 earlier. In use cases like fraud detection where there is constant battle to combat new types of fraud, it is crucial for the companies to boost experimentation speed to avoid losses.

To try out Canso, please visit our website https://www.canso.ai. There is a seamless onboarding flow to get you started in under 30 mins. And you get 1 month free access to test various features.

CansoAI Platform — What’s next?

Canso is working on a much larger initiative to solve the complex decisioning processes that usually constitute the rest 50% of the time of Data Scientists and of Product owners as well. Right from identifying avenues for model improvements and validating the same to deploying and scaling it in production, there are several small to big decisions that are required to be made by tech and business teams that consume a lot of time. For e.g. In classification use cases, optimizing the trade-off between value and revenue loss due to false positives is a crucial component that causes inefficiencies.

We are working on building AI based features that can help both ML & product teams make decisions efficiently and improve model performance easily. Beta for such features will be released in coming months. If you are interested in knowing more and getting early access, please drop us a note here or join the Canso Slack community to connect with the founders.

Open Source Releases

In addition to the feature platform MVP, we’re pleased to open source two very useful frameworks -

K8s Clusters — Framework to easily spin up a Canso compatible K8 cluster. It contains Terraform scripts/modules to provision EKS (Kubernetes on AWS) clusters & related dependencies such as VPCs, Security groups, IAM roles, S3 buckets & RDS Databases. This is intended to provision infra as code in a standardized manner.

Helm Charts — This repo comprises multiple helm charts, some of which such as Grafana, Prometheus, Loki, Airflow are quite commonly used for any Kubernetes cluster & adopted by Data & ML teams. The remaining ones are specific to the Canso platform and its developer agent.

Please check the github repository for more information. And join Slack community to connect with us.