The form has been successfully submitted.
Please find further information in your mailbox.
Innowise has updated a plethora of web apps covering fashion, art, architecture, food, health, and more and leveraged AI capabilities for text-to-image generation & content recommendations.
Our client is a prominent media group producing digital content with a significant presence in Denmark, Norway, Sweden, and Finland. They publish magazines, newspapers, and digital media, covering lifestyle, entertainment, health, and current affairs, free or by subscription.
Detailed information about the client cannot be disclosed under the provisions of the NDA.
As the trend toward digital media consumption continued to rise, the client faced the challenge of keeping pace with the shift. They needed to ensure their digital platforms were not only accessible but also engaging enough to connect more meaningfully with their target audience. With thousands of visitors monthly, they wanted to make their web applications more interactive, visually attractive, and user-friendly, remediate content discrepancies, and improve overall manageability.
Beyond that, they showed interest in implementing artificial intelligence into their workflows to deliver more relevant content and reduce operational costs.
In the first stage, Innowise reviewed the customer’s digital media ecosystem to remediate obvious inconsistencies and find areas for improvement. Besides mitigating errors with navigation, page speed, SEO consistency, content presentation, and more, our project team embarked on Labrador CMS migration. Through “headless CMS” architecture, the content repository and presentation layer are separated, making this platform an ideal solution for modern digital publishers experiencing rapid growth.
Innowise has updated a web application that offers a comprehensive guide to entire homes, covering interior details, architecture, and art. As a leading publication and online platform, this digital media remains a go-to source for innovative architecture in private homes.
We have modernized the web app that provides fresh insights into children’s development and growth. It supports mothers through every phase — from pregnancy to adolescence — making the journey of motherhood more fulfilling.
This digital media has been sourcing, evaluating, and delivering the latest and most crucial health, exercise, beauty, and nutrition updates. Our project team has refactored lifestyle media channels, including articles and features on maintaining a healthy lifestyle, diet advice, exercise tips, and psychological well-being.
This media is a good match for staying informed about royal family updates and the Swedish entertainment scene. For over a decade, the web application has been a reliable source for royal news, eventually evolving into a prominent news outlet for Sweden’s most intriguing celebrities and entertainment personalities, regularly featured on TV.
Since professional photography incurs costly expenses, including skilled photographers, experienced stylists, props, equipment, and studio setups, Innowise suggested developing a novel solution to eliminate the need for manual labor.
Our project team selected StableDiffusionXLand GPT-3.5 to generate high-quality images from text prompts. Initially, we collected parent photos for reference and used LoRA (low-rank adaptation of large language models) to generate realistic pictures. Next, we created a user-friendly text-to-image interface to interact with the model.
AI uses LLM & NLP techniques to understand the text prompt, grasping the request’s content, context, and subtleties. Then, it interprets the features described in the text, such as objects, colors, textures, and spatial relationships, to create real-life pictures based on correlations among textual descriptions and visual elements. If the final output does not meet anticipated expectations, we continuously refine the AI model based on feedback and performance to achieve satisfactory results.
We attained the following results once our ML specialists fine-tuned the image generation workflow based on the prompts.
Example 1: “Steak with garnish, top-down, natural light, on a smooth plate, simple and elegant, captured like a photo taken with a Canon EOS R and 50mm lens in a completely white background with soft shadow, 8k resolution, true texture and detailed photo, high angle.”
Example 2: “Macro photography close-up of mouthwatering lasagna, with layers of perfectly cooked noodles, savory ground beef, and a blend of three gooey, melted cheeses. Add a homemade tomato, meat sauce, and a creamy mixture of ricotta, mozzarella, and Parmesan. Make the sauce using tomato paste, water, sugar, basil leaves, fennel seeds, Italian seasoning, salt, pepper, and fresh parsley. Use a Canon EOS 5D Mark IV and a Canon EF 100mm f/ 2. 8L Macro IS USM lens to capture this indulgent Italian dish’s intricate layers and vibrant colors. Illuminate the scene with warm, soft lighting to accentuate the comforting nature of the dish.”
As our customer faced decreased user engagement, customer retention issues, and a lack of ideas for valuable content, we implemented an AI-driven content recommendation system. It gathers user data, including browsing history, search queries, interactions (like clicks, likes, and shares), purchase history, and demographic information. The AI system uses the collected data to create a profile for each user, encapsulating their preferences, interests, and behavioral patterns.
In the next stage, AI analyzes the user data, combining such algorithms as collaborative filtering, deep learning recommendation machines, and a hybrid method.
Collaborative filtering makes recommendations based on the behavior of other users with similar profiles or preferences. For example, if User A likes certain articles and User B has tastes similar to User A’s, the system might recommend those articles to User B.
The deep learning recommendation approach, in turn, collects vast amounts of data related to user behavior and interactions, including preferences, clicks, searches, likes, and other relevant actions. Then, deep learning models create user profiles and suggest content representations by analyzing collected data. This approach identifies complex patterns that traditional algorithms might miss, allowing for a more nuanced understanding of user preferences.
The hybrid method combines collaborative and deep learning recommendation machines to improve recommendation accuracy and overcome the limitations of individual methods.
Our team ensured the system recognized the user’s preferences and adjusted recommendations based on historical data and current trends to forecast what content would resonate with the target audience.
Front-end
CSS, Next.js, React, Typescript, Labrador CMS
Back-end
Node.js
DE/ML
Python, PyTorch, Keras, NVIDIA TensorRT, NVIDIA DLRM, HuggingFaces, Spacy, Openai API (GPT-3.5), StableDiffusionXL, Docker, Docker Compose, Tensorboard
CI/CD
AWS, Cloudflare, Vercel Currently
Using Agile methodology, we split the project into several stages, greatly enhancing flexibility, communication, and customer satisfaction.
During the iterative discussion throughout the discovery phase, we gained a comprehensive understanding of the client’s requirements and clearly defined the project scope.
In the design phase, our talented UI/UX designers created user stories, customer journey maps, and initial design mockups to enhance user engagement and eliminate existing inconsistencies in web applications. Design sprints facilitated rapid prototyping and feedback gathering, essential for Agile environments.
With two-week sprints, the development stage included daily standups, sprint planning, and retrospectives. Functional components were delivered after each sprint, marking specific milestones. The project team held daily standup and sprint reviews for client demonstrations via Google Meet while handling task prioritization in Jira and maintaining project documentation in Confluence.
2
Product Owners
1
Technical Lead
1
Growth Analyst
1
Scrum Master
2
Back-End Developers
4
Front-End Developers
2
UI/UX Designers
2
ML Developers
1
Cloud Solutions Lead
Innowise modernized the customer’s ecosystem of web applications, delivering more convenience and attractiveness for end users. We migrated the customer’s digital systems to the Labrador CMS, particularly suited for high-traffic digital publications in terms of intuitive interface, ease of use, cost-effectiveness, and functionality. Additionally, we implemented a text-to-image generative AI that converts written descriptions into corresponding images without costly professional photography. Also, we developed an AI-driven content recommendation system that suggests content tailored to the user’s individual preferences, behaviors, and interests.
This resulted in boosted user engagement by suggesting relevant and interesting content without inconsistencies and errors across various digital touch points.
12%
influx in monthly visitors
66%
reduction in professional photography costs
Having received and processed your request, we will get back to you shortly to detail your project needs and sign an NDA to ensure the confidentiality of information.
After examining requirements, our analysts and developers devise a project proposal with the scope of works, team size, time, and cost estimates.
We arrange a meeting with you to discuss the offer and come to an agreement.
We sign a contract and start working on your project as quickly as possible.
© 2007-2024 Innowise. All Rights Reserved.
Privacy Policy. Cookies Policy.
Innowise Sp. z o.o Ul. Rondo Ignacego Daszyńskiego, 2B-22P, 00-843 Warsaw, Poland
By signing up you agree to our Privacy Policy, including the use of cookies and transfer of your personal information.
Thank you!
Your message has been sent.
We’ll process your request and contact you back as soon as possible.
Thank you!
Your message has been sent.
We’ll process your request and contact you back as soon as possible.