Omar Olivares Urrutia

Maxxa Software SpA · Santiago, Chile

Fintech (credit & surety). Own production LLM platform work on the TECH team — governance, evals, observability, and agent tooling for financial products.

• Built LLM governance, evaluation, and observability for production models serving financial products
• Shipped core + enterprise LLM platform features used across internal AI products
• Built Model Context Protocol (MCP) server exposing company data and tools to AI agents and internal teams

Universidad de Talca

Multi-agent LLM platform for SQL education — real-time syntactic, semantic, and logical feedback on student queries.

• Designed multi-agent pipeline for granular SQL feedback (syntax, semantics, logic)
• Benchmarked frontier LLMs (Grok, GPT-4o, Claude, Gemini, DeepSeek) for tutoring quality
• Built eval suite: 19 LLMs × 300 cases with LLM-as-a-Judge + human-annotated ground truth
• Co-authored peer-reviewed paper accepted at DataEd'26 on LLM SQL error detection

Kilo Code · San Francisco, CA / Amsterdam

Early hire (12-person founding team) on Kilo Code, a high-growth open-source coding agent (Cline/Roo-Code lineage).

• Helped launch and scale one of the fastest-growing open-source coding agents (~10K GitHub stars)
• Built OpenRouter integrations and docs that supported #1 model ranking on OpenRouter
• Wrote developer onboarding guides and reference integrations for the agent ecosystem
• Worked with leadership on technical architecture and go-to-market (Sid Sijbrandij's team)

Emergent Mind · North Carolina, US (Remote)

Production RAG for scientific discovery — search and analysis over arXiv-scale corpora.

• Redesigned RAG (chunking + hybrid retrieval + reranking) for 200K+ indexed papers
• Built daily arXiv ingestion/indexing pipelines and W&B evaluation harness
• Shipped FastAPI + Docker services with tracing and reliability tests for production serving

Neura Pod · San Diego, CA (Remote)

Neurotech & AI content; grew channel past 80K YouTube subscribers with multiple 200K+ view videos.

• Translated frontier BCI/AI research into accessible content for a broad technical audience

Centro Nacional de Inteligencia Artificial · Santiago, Chile

Model evaluation for LATAM Spanish-language LLMs.

clicOH · Remote

Backend engineering at logistics startup (RayoApp → clicOH acquisition).

• Backend development and post-acquisition system integration
• Maintained legacy systems while migrating features to the new architecture

Suncast · Las Condes, Chile

Renewable-energy forecasting platform; helped move a monolith toward microservices.

• Refactored services and APIs; optimized database queries for forecasting workloads

Postgraduate Diploma · Centro Nacional de Inteligencia Artificial (CENIA)

• LLM Fine-tuning (LoRA, QLoRA, PEFT)
• Production ML (PyTorch, MLflow, W&B)
• Multimodal Models & RAG Systems
• Ethical AI & Bias Management

Engineer's degree · Universidad Católica del Maule · 5.71/7

• Thesis: Fine-tuned GPT-3.5 for Chile's PAES math exam — accuracy 0.78 → 0.94 (+20%)
• Data Structures & Algorithms, Databases, Distributed Systems, AI/ML

Engineer's degree (coursework) · Universidad de Talca

DataEd'26: International Workshop on Data Systems Education

Empirical evaluation of 19 LLMs on SQL error detection (300 cases): LLM-as-a-Judge + human ground truth. Implications for automated tutoring systems.

Universidad Católica del Maule

Fine-tuned GPT-3.5 for PAES mathematics; accuracy 0.78 → 0.94 (+20%) with targeted optimization and OCR/transcription pipelines.

Automated EPUB generator for Paul Graham's essays — #1 on Hacker News, 800+ GitHub stars

PythonOpen Source

Agentic personal finance assistant with live bank API integration (36h hackathon)

Anthropic APIAgentsFastAPI

Fundación País Digital | EY

Traffic monitoring from CCTV streams (congestion/anomaly detection), designed to scale across many cameras.

Cambridge English