Production agent systems
Building Custom Engine Agents that combine Microsoft Agent Framework, typed tools, Agent Skills, durable memory, streaming, and Microsoft 365 experiences.
Field Notes / About
I am David Carvalho, a Lead AI Engineer and AI Platform Architect focused on production agent systems, applied AI products, and the platforms that make them secure and reliable.
My work connects model orchestration with typed tools, durable memory, identity, data policy, evaluations, and observability. I also build OCR pipelines, retrieval systems, AI coding runtimes, and operator-focused products.
Current professional focus across agent engineering, AI platform architecture, and production delivery.
Building Custom Engine Agents that combine Microsoft Agent Framework, typed tools, Agent Skills, durable memory, streaming, and Microsoft 365 experiences.
Designing the identity, data, model-routing, regional processing, policy, and observability boundaries required to operate agents safely at enterprise scale.
Taking AI systems from evaluation to production with Python, containers, infrastructure as code, secure CI/CD, OpenTelemetry, and measurable release evidence.
The personal systems I use to pressure-test OCR ingestion, embeddings, reranking, research workflows, and operator tooling.
A production logo cleanup and generation bench with upload and prompt flows, worker-backed image processing, brand-pack exports, credits, billing, developer API, and CLI support.
A modern survey product with builder, respondent flows, live analytics, CSV/XLSX/PDF/PPTX exports, Stripe checkout, workspace controls, and enterprise trust surfaces.
A self-hosted Swiss shop for handmade fuzzy wire flowers, combining drops, custom bouquet building, Payload CMS operations, Stripe Checkout, TWINT, and signed order tracking.
A Swiss insurance audit engine with OCR ingestion, TaskIQ-backed processing, pgvector retrieval, configurable embeddings and reranking, and Agno agents for overlap and coverage-gap analysis.
A local-first AI operations system with gateway routing, queueing, memory, governance, and runtime adapters for Codex, Claude, and Gemini.
An AI-native X intelligence toolkit for search, watch, sentiment, trend detection, and polished research reports.
A Polymarket and MT5 runtime with daemon orchestration, monitoring, and operator-focused decision workflows.
The frameworks, data systems, and delivery controls I use to build and operate production AI products.
A quick look at recent case studies, experiments, and product work.