Ivaylo Karapeykov / Product & AI Engineer
I build agentic software and AI products, from the interface to the tools and services behind it. I care about how the whole thing works.
From a first interaction to a finished product.
01Selected work
Five projects, five different problems. Here’s what I worked on, what changed, and why.
Caution7 known vulnerabilities in your version
8 REGISTRIES · WEB, API & MCP · OPEN SOURCE
An open source tool that tells you whether a library is safe to depend on, in the browser or inside your coding agent over MCP.
fewer database round trips
PER-RECORD OPERATIONS → BATCHED TRANSACTIONS
Identified and fixed a critical performance bottleneck in a B2B eSIM distribution platform. Batched DB operations, ~98% fewer round trips.
CLEAR BOUNDARIES. SHARED LOGIC.
Product development across web and mobile for an enterprise, multi-tenant veterinary platform, including an architectural redesign of its diary module.
A considered digital presence for an orthodontic practice. Clear services, patient trust, and a fast, accessible experience.
An industrial services website with a clear path from understanding the work to starting a conversation.
02The toolkit
I work with TypeScript and Python to build AI products, connect tools, and run the services behind them.
Web applications in React and Next.js, mobile apps in React Native. Components, state, forms, and the interactions that tie them together.
APIs, relational data, background jobs, and caching. I also refactor existing applications and track down performance bottlenecks.
Agentic applications, LLM integrations, and MCP servers. I use LangGraph to coordinate workflows, with retrieval, structured outputs, and evaluation around the model.
Choose the tool that fits. Keep the next change straightforward.
03AI engineering
I build software around models, tools, and context. These three architecture examples cover finding answers, extracting data, and connecting tools.
A question + trusted documents
Find relevant context
Build a grounded prompt
Respond with source references
Check relevance & uncertainty
An answer grounded in sources
Retrieve relevant material before generation. Surface the sources, and let the user know when the available context is not enough.
Give it the right context. Check what comes back.
04Behind the work
Slide a tile into the gap to put the pieces in order. You can use the arrow keys, too.
I started with freelance projects, building React and Node.js applications from the database through deployment. Since then, I’ve worked on SaaS products in healthcare, security, e-commerce, and eSIM distribution.
Previously, I worked on the development and evolution of an enterprise veterinary practice management platform across web and React Native. Alongside product development, I redesigned the diary module’s architecture to remove a long-standing maintenance bottleneck.
My work is now fully focused on agentic software and AI products. I work remotely, building applications that connect models, tools, and services with Python, FastAPI, LangGraph, and MCP servers.
Away from the screen, I like travelling and exploring new places.
05The journey
Now
AI product engineering · Remote
Building agentic applications and integrating models with tools, workflows, and services. Working with Python, FastAPI, LangGraph, and MCP servers.
Previous work
Production SaaS · Web & mobile
Building and evolving production products across healthcare, veterinary practice management, eSIM distribution, and security.
2020 — 2021
Freelance · End-to-end delivery
Designed databases, built React and Node.js applications, and owned deployment from the first brief to the finished product.
06What’s next
Tell me what you’re working on and where you need help. We can take it from there.