Before we embarked on the journey to build Tiyaro we ran into multiple customers trying to build AI Solutions. We had customers at different stages of their AI adoption journey. Some were just getting started and had no AI/ML expertise while there were others who had a data science team that had built custom models for their use case. Irrespective, there was a common complaint that we heard from all of them.
- The AI solution space is quite opaque. Where do I find the best models? Who are the best AI vendors for my specific vertical?
- How do I choose an AI solution?
Every developer trying to solve a business problem is faced with these exact same questions. Let's say you are ready to go through the hassle and discovery process to find an AI solution or vendor that works for you. How do you know that you are getting the solution that best suits your needs? Comparing solutions from different vendors means doing a POC (Proof of Concept) with each vendor. Most folks don't have the resources or the time to get into such a prolonged engagement.
The issue is no different if you are simply trying to use off the shelf open source models or models from any one of the many model zoos or vendors who are providing access to their models as a service. A POC to compare models across multiple frameworks, model zoos will take you weeks if not days.
And to top it. The above cycle repeats for every different type of AI/ML problem that you are trying to solve, like computer vision, natural language processing, time series forecasting.
In search of the ML models that can help solve your business problem there are 2 main problems that you will have to solve.
- Find the models that can solve your use case
- Compare the models to find the one that best solves your problem
Find the models that works best for you
Developers who are looking for ML models today have multiple options when it comes to searching for and finding ML models.
The Old Way
The most common places to find the models are
- GitHub Repos of open source models
- Model Zoos - TensorFlowHub, PyTorchHub, ModelZoo,
- Big 3 (Google, Microsoft, Amazon) offering specialized custom models as APIs - Rekognition, Google Vision, Microsoft SaaS
- Startups offering ML models as API - ModelPlace, HuggingFace, NLPCloud
Within this list the way you ‘find’ or ‘search’ for models is
- Searching models on Github is a test of your search skills. No other help there.
- For the other model sources your success depends on how familiar you are with AI/ML. Because a lot of these model zoos are expecting you to search for models by names or by framework or by some data science specific classification of models.
The Tiyaro Way - Explore
Simply search for models in plain English. Give it a try!. The goal of Tiyaro search is to take you from your business case to ml models by simply searching for a model based on your use case. Within seconds you will be presented with a choice of models that are all available as API. Our inventory includes models from open source, other ml vendors and even the Big 3 providers. So you get all the models that you need in one place.
Now that you have a way to find/search/explore models. You need a way to compare the models.
The Old Way
Depending on which model zoo you use. You need to do the following 2 things.
- Onboard all the models that you ‘find’ either on your own infrastructure or access them from the vendors as API.
- Build your comparison test and run it against all these varieties of models.
- Run model inference across all these disparate model frameworks and SaaS model providers.
- Understand the input and output format for each of the models you are evaluating and codify it all in your test suite.
- Collect the results.
- Plot the graphs
- Analyze the results
The Tiyaro Way - Experiments
Tiyaro offers the best of open source models alongside the best SaaS models all on the same platform. We have done all the work to ‘normalize’ the model inference across all these varieties of models, model frameworks, and model providers (open source and SaaS). And not just that, Experiments allows you to ‘compare’ models across different frameworks and vendors without writing a single line of code. Running a comparison on Tiyaro is simply a matter of selecting the models of interest and running the experiment on your dataset. We even have a large variety of public datasets available for your convenience. Once you kick off the experiment you get the results back in a matter of minutes.
Business case to quantified best model in minutes
We are building the ‘AI consumption’ platform. Developers should consume AI as a service just like they consume SMS as a service from Twilio or Payments as a service from Stripe. AI as API is just one piece of our ‘AI consumption platform’. With Tiyaro Explore and Experiments you can easily find the models that fit your use case and instantly compare those models to find the one that best fits your data.