YUAL, Inc.

UK University needed a custom AI learning platform that could support students, lecturers, and administrators from one environment instead of forcing users to switch between separate AI tools and subscriptions.

We designed and developed My Intelligence from the ground up, covering UX/UI design, product design, full-stack development, role-based workflows, admin tools, and multi-model LLM integration.

The platform brings GPT, Claude, Gemini, Llama, and DALL·E into one login.

• Students use Learning Mode for concept explanations, exam preparation, research support, and academic writing.
• Lecturers use a Teaching & Grading suite for lecture drafting, marking workflows, and AI-assisted feedback.
• Admins manage users, academic content, prompts, and platform configuration through a dedicated panel.

The key product layer is the academic workflow system built around 5,000+ prebuilt prompt templates. Instead of giving users a blank chatbot interface, the platform recommends prompts by role, task, and use case, making LLMs easier to use for non-technical students and staff.

The system also stores per-user AI interaction history, helping users continue previous academic work without losing context between sessions.

Technology stack: Next.js, PostgreSQL, and multi-model LLM API integration.

The platform launched with 1,000+ students and staff onboarded from day one, giving the university a production-ready AI-powered EdTech product built around real academic workflows.

Custom multi-vendor marketplace where location was the main product mechanic. Standard marketplace templates could not support the required map-based discovery, seller storefronts, product listings, checkout flows, and compliance-sensitive user journeys.

We covered product strategy, UX/UI design, full-stack development, marketplace architecture, seller onboarding, RBAC, admin tools, and payment flows.

The backend was built with Laravel 12, PHP 8.2, PostgreSQL 15, Redis, and Elasticsearch. The frontend is a server-side rendered Next.js application styled with Tailwind CSS. The platform runs on Google Cloud using Docker and GitLab CI/CD. Payments are handled through CoinPayments for crypto and fiat checkout. The admin panel runs on Laravel Nova.

The map is the primary UX surface. Buyers land on a Google Maps-based interface instead of a standard feed or product list. We built a custom data layer on top of Google Maps API to keep search results in the 20–30ms range while reducing direct API usage and operating costs.

For sellers, the platform makes it possible to create a storefront, publish products, appear on the map, and start selling without a separate website or internal IT team.

At launch, platform onboarded 550+ sellers and supported 23,000+ live, searchable product listings.

A regional fuel distribution company needed to replace daily logistics planning built around Viber group chats, Excel files, phone calls, and disconnected internal systems.

The company operated 100+ tanker trucks, a network of gas stations, and multiple oil depots. Dispatchers, logistics planners, drivers, and management all depended on operational data, but that data was spread across 1C, GPS systems, fuel level sensors, manual documents, and human memory.

We designed and built a centralized logistics operations platform from scratch, creating a single source of truth for fuel distribution planning, fleet visibility, reporting, and document automation.

1. The platform integrates with 1C, AntLogistics GPS, and fuel level sensors at gas stations in real time. Dispatchers can see truck availability, driver assignments, compartment layouts, document status, and live GPS positions without relying on constant phone calls.
    
2. Management receives automated morning reports aggregated from 7 data sources and delivered by 9:00 AM without manual preparation.
    
3. The AI-assisted planning module supports one of the most complex parts of fuel distribution: multi-compartment tanker logic. Each truck can carry up to 5 fuel types across separate compartments. The system recommends truck-to-station assignments, proposes compartment filling configurations, and flags dump stations for leftover capacity.

The goal was not to fully automate the logistics planner’s work. The platform acts as a decision-support system, giving planners faster and more accurate recommendations while keeping final control in human hands.

We also built a transport application generator that replaced a daily 2–6 hour copy-paste process. Master vehicle and driver profiles are automatically mapped onto each supplier’s required document format, allowing a batch of 20 trucks to be prepared in minutes instead of hours.

To support drivers with low digital adoption, we added a Viber-based voice bot for status reporting. This helped the company avoid forcing standard mobile interfaces on a workforce where many drivers were close to retirement age.

The platform was delivered as a PWA, making it accessible across desktop and mobile devices without requiring a separate native app.

Targeted operational outcomes included reducing logistics planning time from 60 to 20 minutes, cutting document generation from 2–6 hours to under 30 minutes, and lowering dispatcher call volume by 30–50%.