# Petru Arakiss

## Summary

Petru Arakiss is an AI systems engineer in Madrid — twenty years in software, in ML since 2015. He designs and ships production AI: retrieval, agent runtimes, guardrails, and evals for teams past the demo. Open to Staff, Principal, and Architect roles, remote across the EU.

I lead current production AI systems where BIFROST, ORVIAN, and Polaris are proof of a broader practice: language models treated as product infrastructure. The work is retrieval quality, context assembly, agent runtimes, guardrails, eval loops, traces, cost control, permissions, and operator interfaces.

## Landing Sections

The home page (https://www.petruarakiss.com/) is structured for clear entity extraction:

1. **Positioning** — production AI engineer & architect; targeting Staff, Principal, AI Architect, Forward Deployed / Solutions, and LLM Platform roles; Madrid / remote-first EU
2. **Current work** — BIFROST (document intelligence), ORVIAN (workflow runtime), Polaris (internal assistant)
3. **Capabilities** — retrieval, agent runtimes, harness engineering, product surfaces
4. **Operating model** — stack, fit criteria, async production ownership
5. **FAQ** — roles, location, technologies, approach vs prompt engineering
6. **Contact** — email and CV

## Author

- Name: Petru Arakiss
- Title: AI Engineering Lead · Production AI Systems
- Location: Madrid, Spain
- Experience: 20 years in software (2006 - present); in machine learning since 2015
- Current role: AI Engineering Lead & Head of Frontend Platform
- Target roles: Staff AI Engineer, Principal AI Engineer, AI Architect, Forward Deployed AI Engineer, LLM Platform Architect, AI Solutions Architect
- Current company context: Atlax360
- Contact: contact@petruarakiss.com

## Main Topics

- AI product engineering
- Harness engineering for coding agents and agentic workflows
- Context engineering, retrieval, and document intelligence
- Evals, traces, browser checks, review loops, and human checkpoints
- Reliability, observability, cost control, and production ownership
- Full-stack systems with TypeScript, Python, FastAPI, Next.js, Postgres, and Redis
- Public OSS proof: gommage, traceframe, nahuali, and vestig

## Published Writing

### Blog Posts

- [Harness engineering is the missing discipline around coding agents](https://www.petruarakiss.com/blog/harness-engineering-for-coding-agents) — ~3551 tokens
  - Category: AI Engineering
  - Published: 2026-04-30
  - Description: OpenAI and Anthropic are converging on the same lesson: coding agents do not become reliable because the prompt is clever. They become reliable when the repo, tools, evals, traces, browser checks, and review loops make their work legible.
  - Raw markdown: https://www.petruarakiss.com/blog/harness-engineering-for-coding-agents/raw

### Key Pages

- [Home](https://www.petruarakiss.com/) - Positioning for Staff / Principal / Architect and Forward Deployed AI roles
- [About](https://www.petruarakiss.com/about) - Background, operating model, and fit
- [Writing](https://www.petruarakiss.com/writing) - Essays on harness engineering and production AI
- [Projects](https://www.petruarakiss.com/projects) - Public proof inventory for current systems and open work

### Agent-facing Files

- [llms.txt](https://www.petruarakiss.com/llms.txt) - This file
- [sitemap.xml](https://www.petruarakiss.com/sitemap.xml) - Standard sitemap
- [feed.xml](https://www.petruarakiss.com/feed.xml) - RSS feed

## Social Profiles

- Twitter/X: https://x.com/petruarakiss
- GitHub: https://github.com/Arakiss
- LinkedIn: https://linkedin.com/in/petruarakiss

## Editorial Focus

This site focuses on AI systems that have to work after the prototype. Content covers:
- Why prompts are only one layer of a larger harness
- Architecture patterns that survive real repositories and production traffic
- Context engineering, retrieval, and document evidence
- Evals, traces, browser checks, review loops, and operator-facing interfaces
- Cost control, reliability, permissions, and failure handling

## Technical Details

- Built with Next.js 16, React 19.2, TypeScript
- Content: MDX blog posts
- Focus: technical depth, production reality, and first-hand engineering judgment
- Updated regularly with production insights

---

Last updated: 2026-06-05