Field notes on reliable LLM systems

Where models
meet reality.

I’m Antonie Chirilus, AI Platform Engineer. I write about running LLMs in production — inference, agents, and the parts nobody demos.

Read essays See projects
Role
AI Platform Engineer
Focus
LLM serving, evals, agents
Based
Bucharest, RO
Stack
  • Python
  • llama.cpp
  • MCP

01  ·  Essays

Notes on the unglamorous
half of agentic AI.

Field notes from production LLM work — what breaks, what holds, and the math that decides which.

02  ·  Selected work

Things I’ve built
in the open.

Projects I build to understand a problem properly — agent infrastructure, retrieval, security and evaluation. Every card links to the code.

All repositories on GitHub

001Production LLM app

AI Interview Prep

AI mock-interview platform. CrewAI agents research the role and build a tailored question bank; an interviewer and an evaluator then run and score the session. Every model call goes through one gateway that owns retries, timeouts and hard spend caps.

  • Python 3.12
  • CrewAI
  • OpenAI
  • NiceGUI
  • SQLAlchemy
View on GitHub

Spec sheet

  • 01One LLM gateway — retries, timeouts, spend ledger
  • 02Spend caps: $0.50 per session, $5 per user per day
  • 03Typed Pydantic outputs for every agent
  • 0450 tests · ruff + mypy --strict in CI
Year
2026

More on GitHub

03  ·  Experience

Where the hours have gone.

  1. 2026.09 — now

    AI Platform Engineer

    Bucharest

    Building the internal platform for serving and evaluating LLMs.

  2. 2025.10 — 2026

    R&D Engineer · AI

    Keysight Technologies · Bucharest

    Built a local AI assistant that lets users configure a networking platform in natural language. On-device inference with llama.cpp.

  3. 2025.06 — 2025.09

    AI Engineer · Intern

    Keysight Technologies

    Production RAG over Keysight’s audit corpus. Function-calling agents on Azure OpenAI, retrieval through Azure AI Search, FastAPI.

  4. 2024.09 — 2025.02

    ML Engineer · Contract

    Roglia SRL · Remote

    Custom AI chatbot for public institutions. CrewAI, LangChain, Voiceflow, Groq + OpenAI inference, Grafana for observability.

  5. 2022 — 2026

    BSc Computer Science

    University of Bucharest

    Before that: several years of mathematical olympiad at the national stage.

04  ·  Correspondence

Reply, disagree, collaborate.

If an essay here was useful, wrong, or worth arguing about — email me. Happy to compare notes on LLM inference, agents and AI platform work. The shorter the email, the faster the reply.