What is our primary use case?
My main use case for Voyage AI is as an AI Automation Engineer, where I build workflow automations on a platform such as make.com. Throughout these months, I have had two use cases where I have particularly used Voyage AI: building a chatbot for the company or creating a resume screener to find the best match according to the job description. Voyage AI's text embedding models are central to this work, and I typically use the free tier models such as Voyage 3 Large or Voyage 4, with Voyage 3 being my primary choice.
When diving into the use cases, I first create a pipeline on make.com and use the text embedder inside it through Voyage AI's API for that particular model. One example of this use case is a chatbot that I built for a company, which was integrated into Slack for the interface. The company could store all documents, including legal documents, and query them whenever they needed clarity or had forgotten something. For example, if I have a document about the company's last year's revenue and I want to query the bot, I can ask what the revenue was for last year and receive that information. This is called a RAG chatbot that I built.
The pipeline involves chunking the documents, where I used OpenAI for spacing through make's internal tools. After chunking, the next important step in the pipeline is embedding, which I leverage through Voyage AI models. After embedding, the vectors are stored in Quadron, making embedding a crucial part of that pipeline taken care of by Voyage AI model.
Out of the two main use cases of the chatbot and the resume screener, both have brought value, but if I have to pick one, I would say the resume screener. The reason being it is a very important task when dealing with a thousand resumes for the recruiting agency. Whenever they receive a new job description, they have to find the best resume, either manually or based on the names saved for each file. Once we feed all those thousand files into the RAG pipeline, where the embedding is handled by Voyage AI model, the text embedding gets vectorized by this model. Whenever a job description comes, that content, primarily in a doc format, also gets embedded using Voyage AI model. Whenever text embedding is needed, it is handled by Voyage AI model, which is the most critical part of the entire pipeline.
I apply Voyage AI through the APIs it provides and try to limit my usage on a daily or weekly basis to avoid running out of the free tier limits. I mainly use Voyage 3 model for the pipeline to ensure it runs properly without excessive credit usage.
What is most valuable?
The best feature Voyage AI offers is its text embedding. The quality of the text embedding is very good and crucial. The core objective of the text embedder model is its contextual understanding; if the document provided is not understood in context, the entire pipeline becomes ineffective. Whenever I ask a question, if the model embedding the resumes or documents is not able to understand the meaning, the answer will not be accurate. This is vital, and I think Voyage AI has excelled in text understanding and contextual comprehension.
The contextual understanding feature helped greatly in the resume screener, where resumes are often one or two pages long and filled with technical terms. It effectively matches those terms with a job description coming from the company, identifying the best match from thousands of resumes. The text embedding does a great job of completely understanding each resume, including technical terms and expectations from the job description. The contextual understanding provided by the text embedder model is excellent.
Voyage AI has positively impacted my work, particularly regarding cost. If I had opted for another model, such as an OpenAI text embedder, the pricing would have been significantly higher, making it quite expensive to use OpenAI for handling thousands or even hundreds of resumes. Although it took time, my experience with Voyage AI on the free tier was superb.
What needs improvement?
Improvement can be made in the documentation aspect, which definitely needs to be updated or enhanced. I felt this need while using Voyage AI. I recognize that it is still growing, which is acceptable for a user like me, but enhancing documentation would greatly help developers or technical personnel.
One more improvement could be native integration. For example, when using Voyage AI through make.com, which is an automation platform I regularly use, integration is possible only through APIs. This requires more time for setup and handling configurations. If Voyage AI had a native integration with make.com, similar to how OpenAI's ChatGPT operates, it would streamline the onboarding process very effectively.
Other improvements needed for Voyage AI include enhanced documentation, native integration, and potentially video tutorials for each model or feature that they add. Short videos could significantly aid developers who might be lost or unfamiliar with updates since staying up to speed can be challenging.
For how long have I used the solution?
I have been working in this field for around eight to twelve months.
What do I think about the stability of the solution?
In my experience, Voyage AI is stable; I have not encountered any issues with instability.
What do I think about the scalability of the solution?
Voyage AI's scalability is commendable; I successfully handled hundreds to thousands of resumes with ease, proving it is quite scalable.
How are customer service and support?
I did not need to contact customer support, as I had no significant issues to warrant this, and the documentation mostly covered initial setups. I found the customer support to be well-regarded based on research from various review platforms.
Which solution did I use previously and why did I switch?
I did not specifically use a different solution for the text embedding model, but in the automation and AI fields, I have explored platforms like OpenAI, Claude, and Gemini. I chose Voyage AI specifically for text embedding due to the provided free tier.
What was our ROI?
I have definitely noticed cost savings. While dealing with thousands of resumes, the process was not completed in one go; it was done in parts. Therefore, the time consumed was slightly more compared to other models, but I had ample time to work without the need to rush, thereby saving on costs while maintaining satisfactory accuracy.
Money saved is the primary metric I can share. The question of fewer employees is not relevant in my situation since I am the only technical person involved, and therefore, it did not affect staffing. While time was consumed, the focus was primarily on saving money, which was accomplished by sticking with Voyage AI's free tier instead of switching to paid models.
Which other solutions did I evaluate?
I evaluated options primarily around OpenAI models, which are costly in comparison to what Voyage AI offers through its free tier, leading me to ultimately stick with Voyage AI.
What other advice do I have?
In addition to everything I have mentioned, I am also using the free tier, which is pretty beneficial.
The best feature Voyage AI offers is its text embedding. The quality of the text embedding is very good and crucial. The core objective of the text embedder model is its contextual understanding; if the document provided is not understood in context, the entire pipeline becomes ineffective. Whenever I ask a question, if the model embedding the resumes or documents is not able to understand the meaning, the answer will not be accurate. This is vital, and I think Voyage AI has excelled in text understanding and contextual comprehension.
My advice for others considering Voyage AI is to prepare for a time-consuming setup, especially when integrating it with automation platforms like make.com. This should be expected before starting. It is essential to understand the documentation but also to seek out video resources online to aid your experience. Overall, it is a solid model, and those interested should familiarize themselves with the documentation and models that Voyage AI offers to relate them to their specific use cases. I would rate my overall experience with Voyage AI as an eight out of ten.
Which deployment model are you using for this solution?
Public Cloud
If public cloud, private cloud, or hybrid cloud, which cloud provider do you use?
Other