Exploring AI Solutions: A Professional Perspective

We participated in several seminars focused on the implementation of proprietary artificial intelligence solutions.

Given the nascent state of the market, our primary objective was to outline a roadmap for AI projects.

Link of the day.
Do you need security compliance for your AI solutions? Here.

Link of the day.
Need an AI estimate? Here.

Established vendors, those with a decade or more of experience in this domain, typically necessitate engagement with their sales division. Following negotiations, a solution architect is assigned to bridge the gap between project requirements and delivery. These architects are seasoned professionals adept at integrating cutting-edge, off-the-shelf solutions.

Prominent vendors include IBM, Accenture, Databricks, Cloudera, and OpenAI. While their solutions may come at a higher price, they offer reliability, particularly if you have a pre-determined budget and the primary risk is ensuring successful delivery irrespective of costs.

Alternatively, you could consider engaging a software development company. We can recommend such firms, which often bring their own expertise and white-label solutions. However, alignment with your requirements may vary. These companies typically operate on an hourly outsourcing model or a fixed-budget project. If you're hesitant about the risk of hiring a team, feel free to reach out to us. We can assist in providing a reliable estimate.

We advise that preparation is key when dealing with AI projects. Participating in training sessions can be beneficial. We've compiled a list of non-traditional vendors that allow you to test solutions with your own team, aiding in negotiations and project scope assessment.

We've experimented with Datastax's solutions, which include vector databases and a UI tool, LangStream. This tool allows you to integrate chat-based interfaces with streams that process data for intelligent consumption. Building a custom AI solution with your own content, such as that on a webpage, takes about a day for a single individual. Some Unix knowledge is necessary.

We've also explored Snowflake's solutions. As a traditional SQL data lake vendor, Snowflake has recently expanded its AI capabilities. Creating an end-to-end solution with Snowflake takes about half a day and requires some SQL and Python knowledge, but a full-stack engineer with Unix skills isn't necessary.

Major cloud providers are another option. We've tested Microsoft Azure's solutions, which offer user-friendly interfaces seamlessly integrated with their own systems. If you're already using Azure, it's a viable choice for building a solution within a few hours.

Amazon Web Services offers a solution called Amazon Bedrock, one of the most comprehensive enterprise solutions with extensive compliance and security features. However, this can slow down integration. If you have a skilled AWS-certified team, they are the right choice.

We recommend assessing where your raw data is stored. If it's in AWS or Azure, their own compute services are the ideal choice. If you have external web content, particularly from media sites, Datastax and Google may be suitable. Snowflake proved useful for integrating with a set of PDF files.

Ultimately, professional organizations may wish to establish an in-house backup system. We built an inference engine using a Dell server and NVidia GPUs, enabling us to determine the minimum cost NVidia hardware required. We'll share more details in a subsequent article.

This article was revised on Jun 24, 2024.