Professional Experience - AutoU (Applied AI Startup)

About AutoU

AutoU is a startup building applied AI solutions for companies — computer vision, AI agents, predictive models, and data platforms — with projects for enterprise clients such as Saint-Gobain, Rocester, Oxiquímica, and retail/fuel chains (plus enterprise demos, such as the pre-sales demo built for Itaú and the demo built to compete in an ongoing Libbs bid/RFP). The environment is fast-paced, with high client exposure and a focus on quality delivery at high velocity.

Work Context

• Projects for large enterprises, all in production
• Core stack: Python (FastAPI, google-genai, LangGraph) in the backend and React 19 + Vite in the frontend — direct LLM integration via the google-genai SDK / Responses API as the main day-to-day approach, alongside LangGraph orchestration
• Multi-cloud infrastructure: AWS (EC2, S3, Lambda, DynamoDB), Google Cloud (Cloud Run, Cloud Run Jobs, Scheduler, Pub/Sub, VMs), Azure
• CI/CD and deploy: GitHub Actions, Docker Compose, automated deployments

Experience at the Company

Fullstack Developer

Period: February 2026 - Present | Remote

24/7 Computer Vision in Production (fuel station chain)

• YOLO pipeline running at the edge (on-site PC consuming RTSP cameras), events consolidated in AWS (S3, Lambda, DynamoDB)
• Agentic pipeline with LangGraph + Gemini (via google-genai) + RAG: analysis of detected occurrences and insight generation with GenAI, plus email/WhatsApp notifications
• Retraining loop: user false-positive feedback becomes retraining dataset — the model improves with use
• Observability implemented by me: Prometheus + Grafana on a dedicated stateful VM, monitoring cost, usage, and infrastructure

Machine Learning and Demand Forecasting (auto-parts chain)

• Predictive algorithm with Prophet (time series) for per-product/per-branch demand forecasting
• Purchase, inter-branch transfer, and substitute product recommendations, integrated with the Sankhya ERP
• Automated daily execution in production

R&D Platform with Bayesian Optimization (chemical industry)

• Formulation platform using BayBE (Merck's open-source framework) to suggest the next formulation to test — fewer physical greenhouse cycles
• Generative AI agent "Colibri" (Gemini via google-genai) that generates and assists with formulas; scientific chat with RAG and source governance (whitelist/blacklist)
• Grafana + Prometheus observability implemented by me: AI cost, consumption, and infrastructure

Savings Replication Pipeline (Saint-Gobain)

• Cloud Run Jobs + Cloud Scheduler with hash-based idempotent synchronization, auditing, and rollback
• FastAPI admin backend; daily execution in production

AI Ticket Triage Assistant (demo for an ongoing Libbs bid/RFP — solo project)

• Full-stack MVP built alone as a demo to compete in the pharma company's bid: public chat with Gemini-powered triage and deterministic rule-based fallback, human handoff, internal portal with ticket timeline
• One application serving two domains (public chat + internal portal) — demo live; bid still in progress

Intelligent Parts Catalog (Rocester — project foundation)

• Foundation participation: architecture with a decoupled AI layer (dependency inversion) and the base of the PDF extraction pipeline with Gemini Vision
• Per-part confidence score with reasoning and bulk human curation; pgvector for semantic search — in production

AutoU Institutional Website (solo project)

• Complete site built alone: SEO-optimized React frontend (blog, cases), FastAPI leads backend, and a custom CMS — live on Azure
• Email integration via Microsoft Graph (Azure app registration): incoming emails are automatically recorded in the CRM funnel, in the same flow as leads captured through the site

B2B People Management Platform (Aura Central — DDD monorepo)

• Notifications feature: dedicated service consuming bounded-context events via Pub/Sub, without coupling domains
• Platform logging system, providing operational visibility in production
• Work within a DDD architecture with 5 bounded contexts and database-per-service

Enterprise Demo Frontend (Itaú — two-person team)

• Pixel-perfect interface from Figma with interactive maps (React 19 + Leaflet), under a short pre-sales deadline

Corporate Integrations

• Sankhya/Microwork ERP via API: automation of reports, quotes, and sales flows
• Slack: real-time notifications; Outlook / Microsoft 365: transactional emails

Technologies Used:

• Python, FastAPI, google-genai / Responses API, LangGraph, Prophet, BayBE
• React 19, Vite, TypeScript
• LLMs (Gemini, Gemini Vision), RAG, pgvector, FAISS, YOLO
• AWS (EC2, S3, Lambda, DynamoDB), Google Cloud (Cloud Run, Cloud Run Jobs, Scheduler, Pub/Sub), Azure
• Prometheus, Grafana
• GitHub Actions, Docker, Docker Compose
• Sankhya/Microwork, Slack API, Microsoft 365 / Outlook

Achievements and Learnings

• Every project I worked on is in production — from 24/7 computer vision to daily data pipelines
• Applied AI with engineering judgment: deterministic fallbacks, human curation, source governance, and retraining loops — never "AI for AI's sake"
• Production observability (Prometheus/Grafana) implemented on my own initiative, including AI cost monitoring
• Direct exposure to enterprise clients (Saint-Gobain) and enterprise pre-sales/bids (Itaú demo and Libbs bid demo) in a fast-paced startup environment
• ML beyond LLMs: time series (Prophet), computer vision (YOLO), and Bayesian optimization (BayBE)