Astra DB, now part of IBM watsonx.data vs. Vertex AI

We've been super happy with Astra DB. It's been extremely well-suited for our vector search needs as described in previous responses. With Astra DB’s high-performance vector search, Maester’s AI dynamically optimizes responses in real-time, adapting to new user interactions without requiring costly retraining cycles.

Incentivized

we used Vertex AI on our automation process the model very useful and working as expected we have implemented in our monitoring phase this very helpful our analysis part. real time response is very effective and actively provide detailed overview about our products.this phase is well suited in our org. this model could not applicable for small level projects why because this model not needed for small level projects and without related resource of ML this model not useful. strictly on non cloud org not suitable means on pram not suitable

Incentivized

• We need to be able to process a lot of data (our biggest clients process hundreds of milions of transactions every month). However, it is not only the amount of data, it is also an unpredictable patterns with spikes occuring at different points of time - something athat Astra is great at.
• Our processing needs to be extremaly fast. Some of our clients use our enrichment in a synchronous way, meaning that any delay in processing is holding up the whole transaction lifecycle and can have a major impact on the client. Astra is very fast.
• A close collaboration with GCP makes our life very easy. All of our technology sits in Google Cloud, so having Astra in there makes it a no-brainer solution for us.

Incentivized

• Vertex AI comes with support for LOTs of LLMs out of the box
• MLOps tools are available that help to standardize operational aspects
• Document AI is an out of the box feature that works just perfectly for our use cases of automating lots to tedious data extraction tasks from images as well as papers

Incentivized

• Need better fine-grained Security options.
• The support team sometimes requires the escalate button pressed on tickets, to get timely responses. I will say, once the ticket is escalated, action is taken.
• They require better documentation on the migration of data. The three primary methods for migrating large data volumes are bulk, Cassandra Data Migrator, and ZDM (Zero Downtime Migration Utility). Over time I have become very familiar will all three of these methods; however, through working with the Services team and the support team, it seemed like we were breaking new ground. I feel if the utilities were better documented and included some examples and/or use cases from large data migrations; this process would have been easier. One lesson learned is you likely need to migrate your application servers to the same cloud provider you host Astra on; otherwise, the latency is too large for latency-sensitive applications.

Incentivized

• Customization of AutoML models - A must needed capability to be able to tweak hyperparameters and also working with different models
• Model Explainability -Providing more comprehensive explanations about how models are utilizing features could be very beneficial
• Model versioning and experiments tracking - Enhancing the versioning capability could be good for end users

Incentivized

It's a great product but suffers with counters. This isn't a deal breaker but lets down what is otherwise a good all round solution

Incentivized

Google is always top notch with their security and user interface performance. We use Google's entire suite in our business anyways, so using Vertex became second nature very quickly. I will say, though, that Google does need to come down on the price somewhat with their token allocation. Also, their UI is very robust, so it does require some time for training to really master it.

Incentivized

Their response time is fast, in case you do not contact them during business hours, they give a very good follow-up to your case. They also facilitate video calls if necessary for debugging.

Incentivized

Graph, search, analytics, administration, developer tooling, and monitoring are all incorporated into a single platform by Astra DB. Mongo Db is a self-managed infrastructure. Astra DB has Wide column store and Mongo DB has Document store. The best thing is that Astra DB operates on Java while Mongo DB operates on C++

Incentivized

We tend to adapt and use the platform that suits the customers needs the best. We return to Vertex AI because it is the most in-depth option out there so we can configure it any which way they want. However, it is not quick to market and constantly changing or updating it's feature-set. This makes it suitable for bigger customers that have the capital and time to spend on a bigger project that is well researched and not quick to market like some of the other options that feel like a light-version of this.

Incentivized

We are well aware of the Cassandra architecture and familiar with the open source tooling that Datastax provides the industry (K8sSandra / Stargate) to scale Cassandra on Kubernetes.

Having prior knowledge of Cassandra / Kubernetes means we know that under the hood Astra is built on infinitely scalable technologies. We trust that the foundations that Astra is built on will scale so we know Astra will scale.

Incentivized

• Better uptime due to the managed service having no outages
• Less technical debt because we don't need to worry about upgrading our Cassandra clusters
• Lower cost on infrastructure as a whole
• Quick and easy to integrate vector search into our tech stack

Incentivized

• It is pay as you go model so it'll save more cost of your org. In our case previously we used to incurred 1-2L/Month now we are reduced it to 80k-1L.
• It'll help you save your model training & model selection time as it provides pre-trained models in autoML.
• It'll help you in terms of Security wherein we can use row level security access to authorized persons.

Incentivized