Hernan M.

I'm Hernan, a backend and AI systems engineer with 18 years writing software. I built OfferToro (now Torox) from line 1 and ran it for four years. It started on a single server on AWS Elastic Beanstalk and grew into a distributed setup on EC2 with S3, ElastiCache and Redis clusters, serving millions of clicks per hour with postbacks and fraud detection. That platform is still running. So when I say a system holds up in production, I was the person on call the times it did not. For the last two years I have applied the same discipline to AI: RAG pipelines, agents with approval steps, and text-to-SQL over real business data. What I actually ship: RAG that cites. Hybrid retrieval over Postgres and pgvector, chunking that respects document structure, and reranking, so the answer points at a page instead of sounding confident. Agents with brakes. Human approval before anything irreversible, typed contracts between steps, and retries that do not repeat the paid work. Text-to-SQL that cannot invent a schema. Read-only access, query validation, parameterized statements only. The backend underneath: Python, Node.js, PHP/Laravel when the code is already there, Postgres, MySQL, Redis, DynamoDB, AWS. Recent work: Atiendia, an assistant builder where a business trains its own assistant (product recommendations, CRM actions gated by email OTP) and installs it with one line of JS. IdeaForge, a pipeline that mines forum conversations for article ideas and runs unattended on job queues with retries. Eureka Graph, a research pipeline that turns a library of academic PDFs into claims with provenance in Neo4j. Top Rated Plus here on Upwork, 100% Job Success, 33 jobs and more than $70K earned. I'm in Uruguay, UTC-3, so I overlap a full working day with clients in the US. I'll tell you what I have not done before you pay for it. If a job needs something outside my experience, you hear it in the first message, not after the invoice. If you need one feature that works against real data, or an audit of an AI system already in production, message me with what you are building.