Osman Vilchez
Computer Scientist · Senior AI Engineer · Applied AI Researcher
Research rigor. Production impact.
I started out building computer vision models before LLMs were the conversation, and I have been obsessed with the same thing ever since: making an intelligent system work outside the demo. Today I design multi-agent systems that process tens of thousands of documents a day, and Politecnico di Milano, where I am doing my MSc in AI, keeps me close to the research that comes next.
- Off-road lover
- Rider
- Adventurer
Production AI systems, end to end
From ambiguous requirements to systems that hold up in production: architecture, evaluation, optimization and operations. Classical ML, computer vision and generative AI, on Azure, GCP and AWS.
Experience
- Jan 2026 – PresentCurrent
Senior AI Engineer
Plain Concepts
- Azure AI Foundry
- Multi-Agent Orchestration
- Hybrid RAG
- LLM Evaluation
- Led the evolution of a document-screening system in three steps: a single LLM classifier, then an agent with tools, then a multi-agent architecture where agents are tools for other agents.
- An orchestrator sends each document only to the specialists it needs (hybrid RAG inside agents, web search, multi-persona review, specialized models), across tens of thousands of documents a day.
- One rule behind every design decision: lower the cost per document without lowering recall. A missed case is the one error we never trade for savings, and every change is measured against an expert-labelled eval set.
- A lesson I keep: a search tool wired in without caching made latency and token use spike, and no unit test saw it coming. Since then, a reviewer agent with testing tools checks latency, tokens and queue impact on every PR, before merge.
- Jul 2024 – Jan 2026
Machine Learning Engineer
Zazmic
- Vertex AI
- BigQuery
- Recommenders
- Forecasting
- Agentic RAG
- Classical ML in production on GCP (Vertex AI, BigQuery): recommendation systems, demand forecasting for inventory restocking, and pattern mining on business data.
- Text-to-SQL systems that turn natural-language questions into queries over structured data.
- Advanced RAG: retrieval optimization, reranking, fine-tuning and retrieval-quality evaluation, including agentic RAG where retrieval is a tool the agent decides when to call.
- Computer vision pipelines, GenAI and multi-agent systems, plus model benchmarking and architecture evaluation to choose the right cloud and AI stack.
- Aug 2022 – Apr 2025
Senior Artificial Intelligence Developer
Arroyo Consulting
- Serverless
- GenAI
- MLOps
- Multi-Agent
- Designed and implemented serverless pipelines and cloud AI services.
- Led internal AI transformation initiatives with GenAI technologies.
- Developed multi-agent orchestration frameworks.
- Accelerated AI adoption through reusable POCs and internal tooling.
- Aug 2023 – PresentCurrent
AI Developer
Quanta Solutions
- LangChain
- RAG
- TTS
- Multi-Agent
- Developed LLM-based applications using LangChain, vector databases, and cloud APIs.
- Implemented TTS, voice cloning, object detection, and image/video generation solutions.
- Designed multi-agent architectures for automation and decision support.
- Nov 2023 – Aug 2024
LLM Consultant
Securitec & YaVendio
- LLM
- Inference Optimization
- Vector Search
- Optimized LLM inference pipelines through caching and parallelization.
- Implemented semantic retrieval systems based on vector databases.
- Mar 2022 – Dec 2022
Data Science Analyst
MiBolsillo App
- Clustering
- Prediction
- Data Analysis
- Developed clustering and user behavior prediction models.
- Statistical analysis and data extraction for product analytics.
Hiring or building something serious?
Open to senior AI engineering roles and to complex projects through Vector Ridge Labs.