How Does AI Development Work? (feat. Dalpha's 6-Step Process)
This article is recommended for
-Those who want to create business impact with custom AI but feel lost about how to adopt it
- Those curious about Dalpha's easy and fast AI service development process
Summary
-At Dalpha, we support AI service development through a total of 6 steps.
(Exploring AI services -> AI consulting -> Demo creation -> Feedback and performance improvement -> Production-service preparation -> Ongoing support)
- By exploring the wide range of products in the Dalpha AI Store and meeting with an AI consultant, you can find the AI solution your company needs.
- With fast demo creation in under two weeks on average and continuous ongoing support after production launch, you can use a company-tailored AI service easily and quickly.
Feeling lost about how to develop company-tailored AI?
The government announced that it would actively support the free use of AI by government ministries, agencies, local governments, and private companies. Since then, interest in AI among companies and institutions has grown even hotter.
For those who want to achieve various business results such as cost savings, improved customer experience, and greater operational efficiency through today's hot topic of AI but feel lost about how to adopt it, we introduce Dalpha's AI service development process.
The 6 Steps of the AI Development Process
Dalpha's AI development process consists of 6 steps in total.
1. Exploring AI services
2. AI consulting
3. Demo creation
4. Feedback and performance improvement
5. Production-service preparation
6. Ongoing support
Let's take a step-by-step look at the process of easily and quickly developing a company-tailored AI service!
AI Development Process 1 - Exploring AI Services
While some corporate clients already have a specific AI service in mind, most are not AI experts, so they tend to struggle with deciding which AI to adopt. They hear that AI keeps advancing, but they feel lost because they don't really know how capable it actually is or which services they can adopt.
Explore which AI services your company needs. The Dalpha AI Storelets you easily explore a wide range of highly useful AI services by industry. the blog and the latest AI case studiesOn the blog and in the latest AI case studies, you can also check out vivid real client adoption cases as well as AI adoption status and results within your own industry.
In fact, many clients tell us that their understanding of AI services grew as they explored the AI Store.
After exploring AI services thoroughly, you can press the inquiry button in the AI Store to inquire about the AI services you're interested in.
Want to build a service that isn't in the AI Store? Don't worry. Dalpha provides custom AI solutions tailored to each client. Even if it's not in the AI Store, we'll build whatever feature you want.
AI Development Process 2 - AI Consulting
When you make an inquiryinquirethrough the AI Store, an AI consultant will guide you through the AI service adoption process on the same day via the contact channel you chose, such as text message or email.
Then, a service adoption meeting is held with a professional consultant who has provided AI consulting to around 650 companies.
In this meeting, the AI consultant sharply defines the problem the client wants to solve and specifies the features needed to solve it.
For example, when a company running an online fashion platform wants to adopt a 'similar product recommendation' feature, we narrow down the development direction with questions like the following.
What business results you expect from the similar product recommendation feature
What types and how many products make up the entire product lineup sold on the platform
What factors should be reflected when recommending similar products
Whether there is a priority order of similarity such as brand/color/material/style
Whether products should be recommended in order of highest product image similarity
Whether other factors such as product name, product category, and price range should also be reflected
How often newly uploaded products each day should be reflected
Whether you want to recommend n products in order of highest similarity for all products, or only recommend products above a certain similarity threshold
After specifying the task with the AI consultant like this and completing additional internal technical review at Dalpha, we begin development of the project in earnest.
AI Development Process 3 - Demo Creation Stage
A dedicated PM and AI Engineer with extensive experience in similar projects are assigned to create a custom AI demo for the client.
For the AI demo, we provide it in one of three formats: a results sheet / a demo page / an APIIf you want to quickly check results for a desired test set, we provide a results sheet; if you want to enter inputs directly and run various tests, we provide a demo page; and if you want to integrate the API into your actual environment for testing, we provide an API.
From the moment a project begins, Dalpha opens a channel where the contacts from both companies can communicate quickly, ensuring smooth communication throughout the project.
It varies by project, but on average it takes a minimum of 5 days and a maximum of about 2 weeks to deliver a demo. For long-term projects with a large development scope, we sometimes set milestones of about two to three months with the client.
AI Development Process 4 - Feedback and Performance Improvement
Once the demo is complete, the dedicated PM delivers it and coordinates a schedule for a feedback meeting to discuss the demo. Once the feedback meeting schedule is confirmed, we send over a feedback document where you can write down your feedback on the demo.
After reviewing the delivered demo thoroughly, clients can write feedback on what they're satisfied with, what needs performance improvement, other contract-related questions, and any additional points they're curious about. The dedicated PM and AI Engineer prepare for the meeting based on the contents of the feedback document written by the client.
In the feedback meeting, we discuss the performance and business value of the delivered demo. For example, a client who reviewed the similar product recommendation demo might give feedback like the following.
Because brand is given priority, products with lower similarity seem to be shown first. It would be great if product image similarity were reflected more strongly.
I'd like products with the same design to be recommended first, even if the product color is different.
For T-shirts, regular T-shirts often get recommended for crop tops because the designs are similar. I'd like only crop tops to be recommended as similar products for crop tops.
When a new product is uploaded, I'd like it to be reflected in the similar product recommendation list in real time. About how much time lag is there until updates are reflected?
If a client wants to see the improved results with their feedback applied, we carry out additional performance improvement work. When additional performance improvement work is done, we hold one more feedback meeting after delivering a second demo. If the feedback can simply be edited and applied, we skip a separate second demo and carry out the improvements while communicating with the client during the production-service preparation stage after the contract.
AI Development Process 5 - Production-Service Preparation
After the feedback meeting, once contract-related discussions wrap up and a subscription contract is signed, we hold a meeting for production-service adoption. In this meeting, we check the technical requirements that need to be agreed upon for production-service adoption. The topics discussed in the production-service adoption meeting are as follows.
Aligning the API spec (deciding input and output formats)
Deciding the DB integration method, e.g. uploading files to an S3 bucket at n-hour intervals each day
Deciding the reflection cycle for updated products, e.g. reflecting new products within 12 hours of upload
If additional performance is needed before production-service adoption, final agreement on the target improvement level
If a service page needs to be developed, aligning the final PRD
Establishing milestones for production-service adoption (confirming the production launch date)
After the production-service adoption meeting, both Dalpha and the client carry out the work for production-service adoption. The timeline varies depending on development needs (e.g. whether a front-end page needs to be developed) and the client's internal circumstances, but typically production-service adoption is completed within a minimum of one week and a maximum of one month.
AI Development Process 6 - Ongoing Support
AI is similar to raising a child. It's important to keep using it and steadily improve its performance over time. Even if you're not 100% satisfied with the AI's current performance, by adopting it first and going through the process of training it, you can complete an AI service that's a perfect fit for your company.
Even after the contract, Dalpha increases the AI value our clients experience through close ongoing support. We optimize the model through periodic retraining on the data that continuously accumulates during the production service, and when the base model's performance is enhanced, we automatically update the model without you needing to make a separate request.
In addition, internally at Dalpha, we conduct availability monitoring for each individual service, and we have a process in place where a person in charge is assigned right away to resolve any issues that arise. You can also continuously check the usage and performance level of the AI solution you're using through the dashboard in your Dalpha My Page.
Want to develop AI tailored to your company...?
So far, we've looked together at the process of easily and quickly developing a company-tailored AI service.
Dalpha currently develops and provides AI solutions to a total of 150 corporate clients, ranging from large enterprises such as KT Commerce, Daehong Communications, and CJ OliveNetworks
to startups such as MyRealTrip, MustIt, and MakeUs (the operator of Dingo), with no distinction
Dalpha currently develops and provides AI solutions to a total of 150 corporate clients.
Feeling lost about what AI to develop and how?
Apply for a custom AI consultation.
With Dalpha, you can develop AI services easily and quickly to create business impact?
Apply for a free AI consultation
Articles worth reading together ?
We saved 30 hours on manual category tagging thanks to AI (feat. Qmarket)
'Dingo (MakeUs)' cut short-form production costs by 35% by adopting short-form generation AI!
Travel platform 'MyRealTrip' achieved work efficiency by adopting manual-task automation AI!
The 10-step guide to AI development—this one article has you covered
Sujung Kim
