Petru Arakiss

20 years in engineering. Production AI after the prototype.

Most AI systems fail for boring reasons.

I lead AI engineering at Atlax360. Right now that means BIFROST, ORVIAN, and Polaris in production. Most of the work is retrieval quality, orchestration failure modes, evals, cost control, and interfaces operators can actually use.

MADRID, SPAIN

Current Work

→Keeping BIFROST, ORVIAN, and Polaris reliable in production
→Working on context engineering, evals, and failure handling

Past work includes BBVA, Santander, Bankinter, El Corte Inglés, Decathlon, Google

Writing

All posts

Socrates, Plato, and the AI in my terminal

6 min read

Socrates distrusted writing. Paul Graham says writing is thinking. The argument feels different when the conversation happens in a terminal.

Projects

All projects

gannicus

NEWOSS

LLM-powered synthetic data generation. Declarative schema API, intelligent caching, batching. Multiple providers: Ollama, SGLang, MLX, vLLM.

TypeScript, Bun, LLM, Ollama

llm-metrics

OSS

Metrics collection for LLMs and AI agents. Track performance, latency, and usage across agents, tools, and API requests.

TypeScript, Bun, Observability

RedWheel

AI-powered penetration testing framework. Web vulnerability scanning with local AI analysis using Python 3.14 free-threading.

Python 3.14, Ollama, Security

ReachKit

Distribution intelligence for outreach opportunities. Reddit scraping with AI analysis and SEO content generation.

TypeScript, Bun, Hono, OpenAI

▋