Antonio – Typescript, Node.js, React

An engineering practice focused on AI and LLM systems for funded startups. Antonio architects and builds each engagement personally, from scoping the problem with the client through to production deployment. The work spans agentic assistants with persistent memory and real tool use, document extraction pipelines over unstructured technical PDFs, spec-driven development tooling for AI coding agents, and permission-aware natural language interfaces sitting over clients' production databases. Every project is delivered solo, so he owns architecture, model selection, evaluation and deployment rather than handing any layer off.

• Built a production AI assistant that maintained a personal knowledge graph of a user's people, projects and preferences using cognee, and took real action across Gmail, Calendar and Notion. Implemented agentic tool use with human-in-the-loop approval and one-tap undo on every action, plus reply drafting in the user's own voice.
• Built a spec-driven development platform where specialised agents generated a full product specification across eight layers, served it to Claude Code through an MCP server pinned to committed versions, and an independent verifier agent checked each delivered piece of work against approved acceptance criteria and logged deviations.
• Designed and built an AI document extraction pipeline that pulled embodied carbon and GWP values out of manufacturer EPDs and technical datasheets for an architecture platform, matching extracted products against the client's existing catalogue.
• Benchmarked models on field-level accuracy against cost per thousand documents, then used a cheaper model for bulk extraction with a stronger model verifying the critical field, and kept the model behind an interface so it could be swapped.
• Grounded all extracted values in EN 15804 carbon databases so the model selected records rather than generating numbers, exposing source, range and manual override on every value.
• Built an AI chat backend answering natural language questions against a client's MySQL data, with ACL filtering so every answer respected the asking user's permissions and project access.
• Built a Slack agent that queried a production database in natural language, respected each user's access level and aggregated results, which the team adopted in place of their internal back office.
• Shipped photo-based food and calorie recognition running in production for paying subscribers on a fitness platform.

It is a software studio building custom products and internal tooling for funded startups and growing companies, and maintains the open-source JMS content management system. Antonio worked across the full stack, from backend architecture to client-facing features.

• Built end-to-end product features across real-time backend and frontend systems.
• Shipped B2B SaaS MVPs, a healthcare platform (GlycanAge), a BNPL payments system (PayWithFour), and AI document-extraction tools.
• Built a TensorFlow bidding optimizer inside the Beeswax real-time bidding pipeline that saved over $100,000 in three months from a single deployment.

It is an AI driven SaaS for behavioral analytics and conversion optimization. Antonio built the product and the engineering team around it from the ground up, architecting the backend and leading development of its core recommendation engine.

• Built a MongoDB and Firestore to SQL converter that moved hot data onto cheaper analytical stores, and designed schemas across MongoDB and PostgreSQL.
• Built HotLeads, a real-time engine on Node.js and TypeScript microservices across Firebase Functions, BigQuery, and S3 that read user behavior live and recommended the next best action.
• Led development of a secure, HIPAA-compliant analytics toolkit for healthcare companies.
• Steered the team to drive innovation, streamline processes, and realize the company vision.

It is a full-service custom software development company in Osijek working with clients across pharma, finance, industrial, and logistics sectors. As a summer intern, I worked on a live client project under senior developer guidance.

• Built and tested features for a .NET based business application as part of a senior-led development team.
• Learned professional software development workflows, including version control, code review, and sprint planning, in an agency setting.
• Contributed to internal documentation to help keep the codebase consistent and maintainable.

A Croatian ecommerce agency specializing in Magento development for international clients. During this one-month internship, I got hands-on exposure to PHP based ecommerce development on a live production codebase.

• Fixed bugs and implemented small features in Magento based ecommerce stores under senior developer supervision.
• Learned Magento's module and theme architecture through direct work on client projects.
• Gained early exposure to ecommerce specific concerns like catalog management and checkout flow.