Mirza Krnjić

(Senior Full-Stack Engineer, TypeScript | Python | AI)

Picture of Mirza Krnjic sitting in the office

About

Senior full-stack engineer with 7+ years shipping web products in TypeScript and Next.js. Most of my work now is Python and AI: Claude-based infrastructure pipelines, LLM features, and retrieval systems that hold up in production rather than only in a demo. I care about clean architecture, fast feedback loops, and code that still makes sense long after launch.

Product Engineering

  • TypeScriptEnd-to-end type safety across app, API, and shared packages, with strict configs and types that catch bugs long before CI does.
  • Next.jsApp Router, server components, and streaming, for SEO-ready interfaces that stay fast on real networks and not just on localhost.
  • Testing & QualityJest, Testing Library, and Playwright wired into fast feedback loops, so refactoring stays cheap instead of terrifying.

Backend & APIs

  • Python & DjangoDjango services for data-heavy and AI workloads, leaning on the ORM, migrations, and admin so business logic stays in one place.
  • Node.jsTyped services built for throughput, clear module boundaries, and failure modes you can actually reason about.
  • API Design & IntegrationREST and streaming APIs, third-party integrations, auth flows, and webhooks that stay reliable under real traffic.

AI Engineering

  • AI Pipelines & InfrastructurePython pipelines that run Claude as infrastructure rather than as a chat box: orchestration across many calls, concurrency and batching, retries with backoff, token and cost budgets, and structured logging so any run can be replayed and audited.
  • LLM IntegrationFeatures built directly on the Claude and OpenAI APIs: tool calling, structured JSON outputs validated against a schema, streaming responses, and context assembly that keeps prompts inside the window as conversations grow.
  • RAG & Vector SearchRetrieval that grounds answers in real data instead of guesses: embedding models, chunking tuned to the source material, vector stores like ChromaDB and FAISS, and reranking so the most relevant passages actually reach the model.
  • Agents & Tool UseMulti-step workflows with LangChain where the model calls real tools and APIs, with narrow tool scopes, bounded retries, and sensible recovery when an individual step fails.
  • Evaluation & Prompt EngineeringPrompt design backed by measurement: evaluation sets, regression runs against known-good outputs, and guardrails, so a change to a prompt or a model version is a decision made on evidence and not on vibes.

Education

2014-2018Bachelor's in Information Technologies, Faculty of Information Technologies Mostar
2018-2020Cisco Networking Academy (CCNA)

Career

2025-presentAI Infrastructure Engineer, Claude pipelines in Python (Freelance / Contractor)
2024-2025Senior Frontend at Apple (Contractor)
2024-2025Full-Stack Developer atEvoila GmbH
2023-2024Senior Software Engineer atCorussoft GmbH
2021-2023Lead Software Engineer atVisiot
2018-2021Senior Frontend Developer (Freelance / Contractor)

Languages

  • English (Fluent)
  • German (Fluent)
  • Bosnian (Native)
© 2026 Mirza Krnjic. All Rights Reserved.