Marcos Santos

ChartLensIQ gives physicians a natural-language interface to patient records, replacing manual chart review with conversation. I built the front end in Next.js and TypeScript and the AI-facing back end in Python and FastAPI.

The core challenge was latency and trust: clinicians will not wait on a spinner, and they will not accept an answer they cannot trace. I implemented streaming responses over Socket.IO, so answers render token-by-token, and structured the LangChain pipeline so every response carries its source context back to the UI.

Alongside the conversational layer, I built the supporting clinical dashboards with React, Chart.js, and Victory, and document generation with React PDF for exportable summaries. Product analytics were instrumented with PostHog to track how physicians actually used the interface and feed that back into iteration.

A civic-engagement platform where organizations manage constituent datasets running to hundreds of thousands of records. I designed and built its AI layer end-to-end. I implemented RAG and semantic search pipelines with LangChain, LangGraph, and the OpenAI API, so the assistant answers using the organization's own data rather than model priors, measurably improving relevance and reducing hallucinations. Retrieval sat on MongoDB, and I rebuilt the query path with compound indexes and normalized search fields, dropping latency by more than 90% across 200,000+ records.

I also led the migration of major platform functionality from web to React Native, delivering a consistent cross-platform experience, and built the geographic layer in React, FastAPI, and the Google Maps API, enabling organizations to visualize and segment their records spatially. Alongside this, I shipped interactive dashboards, reporting, PDF generation, and spreadsheet export, and established end-to-end testing across web and mobile with Jest, React Testing Library, and Cypress.

A ground-up front-end rebuild from Figma designs using Next.js, TypeScript, and Tailwind CSS.

Rather than translating screens one at a time, I first extracted a reusable component library and design tokens, so subsequent screens were composed from existing primitives instead of accumulating one-off styles. The result was a scalable design system that shortened delivery time for later features and kept the product visually consistent as it grew.

The build also covered complex forms, data tables, and document generation using Formik, React Table, and React PDF, as well as Google Maps integrations for location-based views.

Kinvo is one of Brazil's leading investment portfolio applications, giving retail investors a consolidated view of their holdings across brokers and asset classes.

I designed and shipped its Bitcoin Connection feature using React Native and Firebase, allowing users to bring cryptocurrency holdings into a unified portfolio view alongside traditional assets. I also integrated the WebSocket streaming layer, delivering live market prices, portfolio valuations, and financial indicators to mobile.

Keeping the interface responsive under a constant stream of price updates was the main engineering constraint, which I addressed by tuning state management across Redux and MobX and reworking how components subscribed to updates. The feature shipped across both Android and iOS.

A restaurant management platform handling payments, customer engagement, and operational workflows.

I integrated Pagar.me and Braspag for secure multi-brand credit card processing and Talon.One for automated coupon validation and personalized promotional campaigns, building the integration services in Node.js and FastAPI. Back-end services followed Clean Architecture, with role-based access control and secure authentication, built on Node.js and PostgreSQL.

I also built the customer-facing Progressive Web App in React and TypeScript with offline functionality and push notifications, so ordering continued through the unreliable connectivity common on restaurant floors.

Enterprise procurement platforms covering purchasing, supplier management, auctions, and approval workflows, serving more than 40,000 users.

I worked across the stack in React, TypeScript, Node.js, and NestJS on the product side, and built the integration services in Python, FastAPI, SAP Ariba, and GraphQL that synchronize procurement data between enterprise systems. Back-end services were event-driven over RabbitMQ to absorb enterprise-scale workloads.

Most of my impact was on performance. I optimized SQL Server queries and API paths to double back-end request throughput, validated under load with Grafana K6, and cut front-end page load times by roughly 45% through lazy loading, memoization, and rendering optimizations. I also handled production diagnosis across distributed services using Kibana.