Marin – AI agent development, Node.js, PostgreSQL

A B2B platform that collected and analyzed article data, turned it into daily and weekly trend signals, and provided AI chat and content-creation tools. Teams could move from market research to channel-specific content without switching among separate products.

• Built an AI chat interface with React, Next.js, TypeScript, SignalR, and TipTap, including streaming responses, rich-text and Markdown parsing, generated-content flows, and typed API integration.
• Built a visual content studio where users defined templates, guidance, audience, and brand voice, then generated multiple channel-specific content items on a canvas.
• Connected React and Next.js interfaces to typed APIs, Python and FastAPI processing services, PostgreSQL data, and RabbitMQ jobs across the end-to-end product flow.
• Diagnosed context overflow in the article embedding system and added article chunking before embedding, increasing recall by 20% and improving downstream clustering.
• Fine-tuned five to six Qwen 3.5 model versions with LoRA and DoRA, compared four model paths, and improved accuracy by 11% on adapted behavior benchmarks.
• Added internationalization, analytics, caching, prefetching, static generation, SEO, accessibility, responsive behavior, Jest and Playwright tests, and Sentry monitoring to support production releases.

A full-stack SaaS product for managing projects, tasks, calendars, activity, charts, and delivery timelines. Teams could see ownership, progress, and schedule risks in one workspace instead of tracking work across separate tools.

• Built the landing page and full application surfaces with React and Next.js, including project dashboards, task views, activity feeds, calendars, and delivery schedules.
• Connected the frontend to Node.js and Express.js application logic so projects, tasks, calendar entries, activity, and timelines shared consistent data.
• Created more than five operational charts that made task load, project status, and delivery progress easier to scan.
• Implemented Frappe Gantt with drag-and-drop task management so teams could adjust schedules directly on the timeline.
• Translated design files into responsive HTML and CSS while keeping dense project data usable on smaller screens.

A full-stack AI product that analyzes text, audio, and video, then maps the content to predicted brain-region activation and likely emotional responses.

• Built the product as one system with a Python backend and runtime plus a Next.js frontend.
• Connected text, audio, and video inputs to Tribe V2 and DeepSeek analysis.
• Turned model output into a user-facing view of predicted brain activity and emotional response.
• Built the complete flow from input and processing to analysis and the final result screen.

A knowledge extraction system built to make long research tasks easier to structure, audit, and resume. The system uses LLM workflows, embeddings, indexing, phase routing, saved runtime state, and evidence-backed outputs to turn messy research material into structured rules that can be reviewed later.

• Built the extraction and rulebook-generation flow with LLMs, embeddings, HNSW-style indexing, and taxonomy-based phase routing.
• Added resumable runtime state so interrupted runs could continue without restarting the full process.
• Structured specialist handoffs and reusable execution phases for multi-agent processing.
• Added persistent artifacts so prompts, rules, state, and final outputs could be inspected after a run.

A bilingual corporate website for a Croatian trade and logistics business, covering food ingredients, ferry transport, paper, rentals, and real estate. International customers could understand the full service range and reach the right business contact without working through a generic brochure.

• Built the responsive corporate portfolio with React, Next.js, TypeScript, semantic HTML, and CSS from the service structure through production delivery.
• Implemented Croatian and English pages with next-intl, including language-aware routing and switching for local and international visitors.
• Structured five distinct business areas into focused service pages with consistent navigation, partner information, and clear contact paths.
• Added page-level SEO metadata and responsive layouts so the service catalogue remained discoverable and readable across desktop and mobile devices.

A website-audit marketplace where business owners could compare audit packages, request a review, and work with specialist auditors. Customers could understand and buy design, SEO, security, performance, responsiveness, and accessibility work without knowing the technical details beforehand.

• Built the responsive marketplace with React, Next.js, JavaScript, HTML, and CSS, covering service discovery, package comparison, account access, and auditor onboarding.
• Created Basic, Standard, Premium, and custom-plan interfaces so customers could compare design, SEO, security, performance, responsiveness, and accessibility coverage.
• Implemented registration, login, and auditor-application routes to support both sides of the marketplace.
• Structured service, pricing, FAQ, about, and contact pages into consistent desktop and mobile navigation.
• Wrote clear product flows and calls to action that guided visitors from the landing page to a suitable audit plan or specialist application.

A subscription platform where customers could compare professional traders, create accounts, pay for access, manage plans, and connect exchange accounts for copy trading. For customers, that meant one place to discover strategies and automate access; for the operator, it created a scalable subscription business.

• Led end-to-end delivery across product architecture, implementation, and release, coordinating the client, designer, support team, another developer, and the traders represented on the platform.
• Built the React and Next.js application, MongoDB data layer, authentication, subscriptions, payments, and email notifications, then shipped the first release in six weeks against an eight-week estimate.
• Added Google OAuth to registration and login, increasing account creation by 30%.
• Implemented dark mode and in-app messaging, increasing user-feedback signals by 75% and 60% respectively.
• Delivered the commercial platform that generated $30K in its first two months.

A responsive digital-asset banking application where customers could review balances, earn interest, borrow funds, exchange assets, and manage account activity. The focus was making complex financial actions understandable and usable on desktop and mobile.

• Delivered more than 20 React and Redux product screens across dashboards, wallets, deposits, withdrawals, earning, borrowing, repayment, asset conversion, transactions, and account settings.
• Modeled wallet behavior with Redux Thunk, including balance changes, asset holding and redemption, collateralized borrowing, repayment, and conversions between digital assets.
• Connected the product to the Binance REST API with Axios, bringing live BTC, ETH, and BNB prices into the conversion experience.
• Implemented authentication state and protected routing for registration, login, two-factor authentication, KYC, security settings, and account-activity views.
• Built sortable transaction history, CSV export, Recharts price charts, and QR-code security setup so customers could review and manage financial activity without leaving the dashboard.
• Reworked more than 20 routes for dedicated desktop and mobile experiences and cut page-load time by 10% through WebP conversion and image optimization.
• Added Google Analytics to give the client visibility into product usage and navigation patterns.