mscope engaged with DoiT’s expert services to help consult on and address its machine learning needs. As mscope lacked expertise in this specialization, DoiT provided guidance and solutions in various areas.

Firstly, DoiT helped mscope with budgeting and estimating costs for the necessary AWS services. This ensured that mscope could plan its financial resources effectively.

Secondly, DoiT designed a solution for fine-tuning transformers-based models such as Mistral 7B , backed by HuggingFace framework. DoiT engineers created an architecture and workflow that met mscope’s specific requirements.

In the current setup, mscope uses a long-running EMR cluster with several defined steps. The first step involves cleaning and normalizing the web-scraped and RDS data. After this, they create a prompt in the next step. Next, mscope uses the generated prompts to extract relevant data by invoking the SageMaker Jumpstart endpoint in the subsequent step.

But, this architecture was quite expensive and complex to maintain. Therefore, DoiT recommended that mscope opt for sagemaker batch transform to execute these different steps and invoke the sagemaker model asynchronously. Additionally, they could orchestrate these steps together with Step functions.

The proposed architecture is as follows:

In addition, DoiT provided detailed cost estimates for the proposed architecture, helping mscope plan its budget effectively.

Finally, DoiT’s relationship with AWS allowed it to get early access to Amazon Bedrock on mscope’s behalf, which enabled the company to experiment with the latest LLMs at lower costs.

Overall, DoiT’s unrivaled expertise and guidance provided mscope with a customized machine learning roadmap. This roadmap covered data preparation, model fine-tuning, deployment, and budgeting. With DoiT’s assistance, mscope was able to leverage machine learning effectively and create value from its data.