TextKernel Explores Scaling and Affordability of Generative AI for CV Parsing
Firemind partnered with TextKernel, a leader in machine intelligence for HR and staffing, to explore how generative AI could be scaled and become cost effective for CV parsing. The project resulted in a proof-of-concept solution that demonstrated the feasibility of using large language models hosted on AWS Bedrock.
TextKernel Explores Scaling and Affordability of Generative AI for CV Parsing
Firemind partnered with TextKernel, a leader in machine intelligence for HR and staffing, to explore how generative AI could be scaled and become cost effective for CV parsing. The project resulted in a proof-of-concept solution that demonstrated the feasibility of using large language models hosted on AWS Bedrock.
At a glance
TextKernel is a global leader in providing artificial intelligence technology solutions to over 2,500 corporate and staffing organisations worldwide. Their expertise lies in delivering industry-leading multilingual parsing, semantic search and match, and labour market intelligence solutions to companies across multiple sectors.
Challenge
TextKernel had been using generative AI via OpenAI, but wanted to see how Amazon Bedrock could be leveraged to significantly improve the speed, accuracy, and scalability of their CV parsing capabilities, while also reducing the associated costs. The key objectives were to create a scalable, cost-effective solution that could seamlessly handle TextKernel's high-volume CV processing needs, while enhancing the efficiency and flexibility of their information extraction capabilities through the use of advanced generative AI models hosted on the AWS infrastructure.
Solution
Firemind proposed a solution that leveraged AWS Bedrock, to create a defined chain-of-thought process. This process used prompts provided by TextKernel to extract key information from CV data, which was then saved as separate fields in DynamoDB.
Services Used
AWS Bedrock
Amazon S3
AWS Lambda
AWS Step Functions
Amazon DynamoDB
Outcomes
40% cost reduction when using Amazon Bedrock.
2x the speed in comparison to OpenAI models.
Business challenges
Enhancing CV parsing at scale
TextKernel sought a solution to a scalable and cost-effective use of Generative AI for their CV parsing process. The primary objective was to explore how the advanced language models available on the Amazon Bedrock platform could be harnessed to enhance their CV parsing capabilities, while also delivering substantial cost savings and scalability benefits in comparison to OpenAI.
By partnering with Firemind to develop a proof-of-concept solution on Amazon Bedrock, TextKernel aimed to unlock the potential of a cost-effective use of generative AI in their core CV parsing operations.
“In the first half of 2024, Firemind supported us in exploring and evaluating the cost-effectiveness of AWS Bedrock Large Language Models (LLMs) for our information extraction tasks on larger documents, while keeping or improving the quality. During this proof-of-concept, Firemind explored multiple LLMs and various usage strategies, balancing token usage and latency.
The outcomes were great: specific AWS Bedrock LLMs delivered a 40% cost reduction and twice the speed (at similar quality levels), compared to our current third-party LLM provider. Firemind's expertise and comprehensive approach have unlocked new technical opportunities for us that are both scalable and budget friendly.”
Mihai Rotaru, Head of Research and Development — TextKernel
Solution
Harnessing the Power of Generative AI on AWS
To address TextKernel’s challenge, Firemind proposed a solution that leveraged the capabilities of Amazon Bedrock, a fully managed service that offers a choice of high-performing large language models. The team developed a defined chain-of-thought process that used prompts provided by TextKernel to extract key information from CVs, which was then stored in an Amazon DynamoDB database. This approach aimed to streamline the prompt engineering logic and validate the scaling and cost of generative AI for enhancing TextKernel’s CV parsing capabilities.
The solution utilised other AWS services, such as Amazon S3 for data ingestion, AWS Lambda for data processing, and AWS Step Functions for orchestration. By harnessing the power of these AWS technologies, Firemind sought to create a scalable and flexible system that could handle the high volume of CVs TextKernel processes.
Scalability and cost effectiveness
The most affordable of the performant Amazon Bedrock models produced results with 12% to 57% savings compared to the benchmark cost set for the project, yielding potential significant cost savings over OpenAI.
The most performant LLM achieved an average max time of 3.6 seconds, which is 55% of the total time of the 6.5 second benchmark (almost twice as fast as OpenAI).
Model Spotlight
Claude 3 Haiku
We chose Claude 3 Haiku for this project due to its exceptional performance in text extraction and rapid processing times. Haiku was the ideal fit for handling large volumes of CVs because it offered the lowest latency among the Claude 3 models, processing tasks in half the time compared to other models.
Its minimal token usage further enhanced cost efficiency, allowing TextKernel to maintain a high level of precision in resume parsing without inflating operational costs.
Haiku’s strength in delivering consistent output formats, particularly JSON, ensured that the data extraction was both reliable and fast, meeting TextKernel’s needs for scalability and affordability.
Why Firemind
As an all-in AWS partner with deep expertise in data, machine learning, and generative AI, Firemind was the ideal choice to help TextKernel address their need for a scalable and cost-effective solution to enhance their CV parsing capabilities.
Firemind’s extensive experience in leveraging the power of AWS services, including the advanced language models available on the Amazon Bedrock platform, made us well-positioned to develop a proof-of-concept solution that could seamlessly handle TextKernel’s high-volume CV processing requirements.
By partnering with Firemind, TextKernel gained access to the latest advancements in generative AI technology and the ability to explore cost-optimisation and scalability via a broader range of LLMs enabling them to maximise the return on their investment. Firemind’s comprehensive understanding of AWS infrastructure and pricing models allowed us to architect a solution that would deliver notable cost savings, while maintaining the speed, accuracy, and flexibility needed to support TextKernel’s global customer base.
Get in touch
Want to learn more?
Seen a specific case study or insight and want to learn more? Or thinking about your next project? Drop us a message below!