Improve your Business with Artificial Intelligence
PNWSoft is experienced at creating applications using machine learning. We have a proven track record with it. Call us to find out if you have a good use case for using artificial intelligence in your business. If you have the proper use case, we can enhance your business process with a machine learning model.
Use our 28 years of experience to help your business or department succeed. From Fortune 100 companies to startups, we have a successful track record.
Find out more, and talk to us today to see how we can help!
Benefits
- AI writes the code. Experienced developers direct it, and catch it when it goes off track.
- Rapid software releases. Get value as quickly as possible.
- No marketing BS. Talk directly to development leads.
- Hybrid (US & Ukrainian) developers with competitive pricing.
- We focus on your business goals, and design a solution for you.
- Experts at taking over failing projects.
Machine Learning That Earns Its Cost
Most businesses asking about AI do not need a model trained from scratch, and some do not need machine learning at all. The useful first conversation is about which category your problem falls into, because the three answers have wildly different price tags: a rules engine you already could have written, a large language model driven by a carefully built prompt, or a custom model trained on your own labeled data.
We will tell you which one applies. Our explanation of what machine learning actually does and our checklist for evaluating a candidate process are written so you can work most of that out before spending anything.
Where it works well
- Image and object detection. Inspecting welds, finding defects, counting or locating objects in photos and video.
- Classification and filtering. Routing incoming content, flagging inappropriate material, sorting documents.
- Fraud and anomaly detection. Where transaction volume is far past what people can review by hand.
- Summarization and extraction. Turning unstructured text into fields your systems can use.
Custom models and language models
For a custom model, the hard part is almost never the algorithm. It is the labeled data. If you do not have enough examples of both the inputs and the correct outputs, no amount of modeling will save the project, and finding that out early is the cheapest thing we can do for you. Our case study on machine learning in a mobile application covers what that looks like in practice, including running a model on the device with Core ML.
Large language models are a different tool. They are not trained for your project, they are applied to it, and the skill is in prompt construction and in feeding them the right subset of your data. They also make confident mistakes, which means the process around the model matters as much as the model. We used one to help a customer reconcile vendor invoices against customer bills, taking a task that consumed about thirty hours a week down to two, and the build took under a week.
How the code gets written
The application around your model is built the way we build everything else now: AI writes the implementation, our developers decide what gets built and correct it when its suggestions stop making sense. Using AI to build AI systems is less novel than it sounds, and the judgement about whether a model's output is good enough to act on is still the human part.
Security gets the same treatment. Systems that accept uploaded images, documents or free text for a model to process are handing an outsider a direct route into your processing pipeline, so nothing ships without being tested against an AI attacker first. Why that is now table stakes.
How we approach it
Small first. A narrow proof on your real data tells you whether the accuracy is good enough to act on, before anyone commits to production integration. If it is not, you have spent a small fraction of the budget to find out. If it is, we build the surrounding application, because a model that no system can call is not worth much.
What Can AI Really Do?
Artificial Intelligence (AI) is able to enhance a lot of business processes, and enables some new ones. But there are many misconceptions about it. Computers do not have any real intelligence, and
AI doesn't mimic human intelligence. A better name for it is machine learning, which accurately describes how it works.
All machine learning applications start with a model, which has
been trained on numerous examples of both the inputs and desired outputs of some knowledge area. Then when the model is shown a new set of inputs, it is able to determine the corresponding set of expected outputs.
The classic example is image recognition. For example, if you train your model on pictures of airplanes, it will be able to identify any airplanes that it sees. It will not be able to identify spaceships
or birds, unless it was trained on those as well. Find out more details about artificial intelligence applications, or read our case study
about a specific example. And use our checklist to see if your business process or application is a good candidate.
Large Language Models (LLM's) are huge models typically trained on billions or trillions of pieces of data and cost 100's of millions to generate. They are not something that is trained up for a project, but with their
broad based dataset, they can be used with specific subsets of data. Typically the skill involved in using them is in formatting the prompt and data input to work with the LLM. This can be tricky, and is typically an iterative approach.
LLM's often make silly mistakes, and the prompt needs to be extremely specific in many cases to produce useful output. For example, we used a LLM (ChatGPT - but any could work) to help a business reconcile invoices from vendors and bills
to customers. This process used to take about 30 hours a week, and using the LLM, allowed us to reduce the time to 2 hours a week. It took less than a work-week to create the process for the customer.
These types of applications are very suitable for machine learning:
- Image recognition. Do you need to inspect thousands of spot welds, or find holes in a membrane? Machine learning can easily automate that type of process.
- Summarization. Do you have product reviews and need to generate a summary? Machine learning is able to summarize content found in many input responses.
- Fraud. If there is a business process that generates thousands or millions of transactions, it isn't feasible to have a person check them all. But a custom trained model would be able to detect fraud similar to what was found in the past.
- Content Filtering. If you need to weed out inappropriate content (either images or text), machine learning is able to automatically filter incoming content to determine if it is appropriate or not.
- Object detection. Custom models can have a extremely high recognition rate, allowing you to automatically find items in an image or video.
Examples of Our Work
As demonstrated from our body of work, we can make or complete applications, services, websites, mobile apps, or optimize processes for your business.
Do you want to create a new app/website for your business or improve your existing software development process? We have a proven track record.
Some Feedback From Our Customers
Please Contact Us
We would love to talk with you about your project. Give us a call, or fill out the form below and a development lead will call you back. You will not be handed to a salesperson.
Additional Info
Thank you very much for contacting us. We will respond as soon as we can.