How I built an AI Kanban platform with FastAPI, RAG and agents
Architecture notes from Lavc Systems: FastAPI backend, React UI, local LLMs, RAG, vector memory, agents, WebSocket observability, and task orchestration.
Patrick Araujo Engineering Blog
Backend engineering, AI systems, automation, API integrations, cloud architecture, security, and software delivery notes by Patrick Araujo.
Latest posts
How I structure API integrations with Python, SQL and GitHub Actions
A concise pattern for API integrations: credential boundary, raw ingestion, idempotent identifiers, scheduled jobs, SQL modeling, and operational reporting.
Building FastAPI backends with logs, queues and traceability
FastAPI backends become easier to operate when requests, jobs, queues, task state, and logs are designed as one traceable workflow.
How I use LLMs in internal systems without letting AI control critical decisions
LLMs can summarize, retrieve, draft, and suggest, but critical actions need policy, permissions, audit trails, and deterministic execution paths.
Automating operational reports with APIs, ETL and SQL
A practical reporting pipeline starts with reliable API collection, keeps raw data auditable, and moves business logic into SQL models that can evolve.
AI For Biodiversity Monitoring: Cameras, Sensors, And Cloud Pipelines
How to build biodiversity monitoring pipelines with camera traps, acoustic and environmental sensors, edge filtering, cloud storage, AI labels, human review, dashboards, and conservation alerts.
Anti-Fraud Graphs For Regional Fintech
How regional fintech teams can model accounts, devices, documents, phone numbers, payment keys, beneficiaries, transfers, and temporal links as an explainable anti-fraud graph.
Cloud Regions, Data Residency, And Latency In Latin America
How to choose cloud regions in Latin America by measuring user latency, service availability, data location, international transfers, resilience, egress, and operational complexity.
Digital Public Services And GovTech APIs In Latin America
How to design digital public services around identity, forms, evidence, payments, status, notifications, interoperability, accessibility, and auditable APIs.
EU AI Act As A Software Architecture Problem
How to translate the EU AI Act into a software control plane for system inventory, role classification, data governance, technical documentation, logging, human oversight, release gates, monitoring, and incidents.
Field Service Automation For Connectivity Infrastructure
How to automate connectivity field service with network incidents, trouble tickets, work orders, resource inventory, technician routing, SLA clocks, offline mobile forms, evidence, and closure verification.
NIS2 For Developers: Turning Cybersecurity Regulation Into Backend Requirements
How to translate NIS2 into backend requirements for asset inventory, risk management, incident reporting, continuity, supply-chain security, vulnerability handling, access control, evidence, and management accountability.
Building Reliable APIs For High-Variance Networks
How to build APIs for unstable mobile and remote networks with deadlines, retries, exponential backoff, jitter, idempotency, caching, compression, pagination, resumable transfer, and graceful degradation.
Smart Mining And Industrial Telemetry In Latin America
How to design smart-mining telemetry in Latin America with vibration, temperature, pressure, safety events, edge buffering, OPC UA, MQTT Sparkplug, predictive maintenance, industrial cybersecurity, and integrated operations.
Agricultural IoT With ESP32: Soil Sensors, Irrigation, And Predictive Alerts
How to build agricultural IoT with ESP32 nodes, calibrated soil-moisture sensors, offline irrigation control, flow and pressure feedback, MQTT or LoRaWAN telemetry, evapotranspiration models, dashboards, predictive alerts, OTA, and safety interlocks.
AI Customer Care In Telecom: From Call Logs To Resolution
How telecom operators can turn calls, chats, tickets, network events, billing data, and customer history into an AI-assisted resolution pipeline with diagnosis, safe tools, human handoff, verification, and auditability.
Brazil's 5G And AI Opportunity For Developers
What developers can build around Brazilian operators by combining Open Gateway APIs, 5G, IoT, edge processing, customer-care AI, service automation, OSS/BSS integration, observability, consent, and multi-operator architecture.
ERP Integration In Latin America: Omie, SIGE, Zoho, Sheets, And SQL
How to design reliable Latin American ERP integration across Omie, SIGE Cloud, Zoho Creator, Google Sheets, and SQL using raw ingestion, canonical models, identity mapping, incremental sync, idempotent writes, reconciliation, and observability.
Low-Cost Observability For Latin American SMEs
A practical low-cost observability architecture for Latin American SMEs using structured logs, Prometheus metrics, cron monitoring, synthetic checks, sampled traces, actionable alerts, retention tiers, and cost controls.
Spanish-Portuguese-English AI Support Bots
How to design multilingual AI support bots for Latin America with language and locale detection, market-aware RAG, contextual localization, code-switching, grounded answers, tool controls, human handoff, and per-language evaluation.
Nearshore Engineering: API Design Across US And Latin America Teams
How US and Latin American nearshore teams can design and deliver APIs with timezone overlap, contract-first specifications, localization rules, async reviews, CI gates, clear ownership, incident handoffs, and outcome-based delivery metrics.
The 32 Percent Usage Gap: Building Offline-First Systems
How to build offline-first mobile and web systems with local databases, sync queues, idempotent operations, explicit conflict resolution, low-bandwidth UX, bounded retries, secure local storage, and observable synchronization.
Pix, Open Finance, And Reconciliation Pipelines In Brazil
How to design event-driven Pix and Open Finance payment pipelines with idempotency, consent, settlement evidence, double-entry ledgers, refunds, fraud controls, multi-layer reconciliation, and operational observability.
Satellite Direct-To-Device And Resilient Backend Design
How to design resilient backends for satellite direct-to-device and non-terrestrial networks with connectivity fallback, store-and-forward queues, message expiry, idempotency, delayed events, local buffering, degraded modes, and secure commands.
WhatsApp Commerce Automation: CRM, Payments, Logistics, And Human Handoff
How to design WhatsApp commerce automation with Cloud API webhooks, consent, CRM identity, order state, payment confirmation, logistics events, AI assistance, human handoff, idempotency, and observability.
AI Agents In SaaS Admin Panels
A backend architecture for AI agents acting safely on SaaS accounts, invoices, support tickets, and reports through delegated authorization, approvals, durable workflows, and audit evidence.
AI-Assisted Incident Response Runbooks
How to design AI-assisted incident response runbooks that preserve evidence, summarize telemetry, propose containment, open remediation pull requests, require approvals, and verify recovery.
The API Gateway Becomes An Agent Gateway
How gateways evolve for agent traffic with protocol-aware discovery, tool registries, delegated authorization, semantic policy, token and action budgets, event logs, and result verification.
Confidential Computing For Multi-Tenant SaaS Data
How multi-tenant SaaS platforms can use trusted execution environments, remote attestation, policy-bound key release, and controlled outputs to protect sensitive analytics and AI data in use.
Data Center Energy As A Software Architecture Constraint
How power capacity, energy efficiency, carbon intensity, water, accelerator availability, latency, and data residency should shape AI infrastructure, caching, scheduling, and workload placement.
Data Provenance For RAG Systems
How RAG systems should preserve document source, creator, version, license, ACL, parser, chunk, embedding, index, retrieval, prompt, claim, and citation lineage.
Latin America 5G Architecture: From Coverage To Enterprise Use Cases
How Latin American enterprises can turn 5G coverage into operational systems by combining public or private mobile networks, local breakout, edge computing, event pipelines, cloud services, network APIs, dashboards, and measurable outcomes.
Digital Trust In Latin America: Fraud, Identity, And Payment APIs
How to design real-time digital trust systems for Latin America by combining identity proofing, passkeys, device and telco signals, behavioral data, payment context, FAPI-secured APIs, fraud decisions, human review, and recovery.
Model Evaluation Pipelines For Product Teams
How product teams can build continuous model and AI system evaluation pipelines with versioned datasets, deterministic tests, calibrated judges, human review, release gates, and production feedback.
Secure Local LLMs For Sensitive Developer Workflows
How to run local LLMs for sensitive source code with explicit trust boundaries, controlled egress, verified model artifacts, ACL-aware RAG, sandboxed tools, secret protection, and hybrid routing.
AI Agent Standards And The New Backend Boundary
How NIST's AI Agent Standards Initiative, MCP, A2A, OpenAPI, identity, delegation, and auditability are reshaping the backend boundary for interoperable agents.
AI Compute Cost Engineering: Tokens, GPUs, Caches, And Queue Backpressure
How to engineer AI systems around request budgets, token costs, GPU utilization, prefix caches, dynamic batching, bounded queues, backpressure, and graceful degradation.
AI-Native Development Platforms: The CI/CD Pipeline After Coding Agents
How CI/CD platforms must evolve when coding agents generate changes: executable specs, tests, security gates, provenance, attestations, review, and controlled delivery.
AI Security Platforms: From Tool Sprawl To Control Plane
How an AI security control plane joins model inventory, prompts, data flows, identities, permissions, evaluations, runtime logs, vendor risk, incidents, and governance.
Cloud Cost Observability For AI Workloads
How to allocate and observe AI workload costs across inference, embeddings, vector databases, storage, egress, scheduled jobs, tenants, features, and business outcomes.
Non-Human Identity: Service Accounts, Agents, Bots, And Audit Trails
How to choose and govern API keys, OAuth clients, service accounts, workload identities, bots, and AI agent identities with reconstructable audit trails.
Physical AI In Warehouses And Robotics: Backend Systems Behind Autonomous Machines
How backend systems coordinate physical AI in warehouses through missions, fleet adapters, traffic, telemetry, recovery, safety boundaries, interoperability, and simulation.
Post-Quantum Readiness For Backend Engineers
A practical post-quantum readiness guide for backend engineers covering cryptographic inventory, TLS dependencies, token lifetimes, signatures, vendors, testing, and migration.
Preemptive Cybersecurity: Detecting Risk Before The Incident Ticket
How predictive signals, CISA KEV, EPSS, threat intelligence, anomaly detection, attack-path context, and bounded automated containment create preemptive security pipelines.
AI Agent Security: Why Prompt Injection Is An Authorization Problem
Why prompt injection becomes a confused-deputy authorization failure when AI agents can use tools, access data, and act with delegated authority.
Software Supply Chain Provenance For AI-Generated Pull Requests
How AI-generated pull requests should preserve task, agent, model, tool, diff, review, commit, SBOM, build provenance, signatures, artifact, and deployment traceability.
TypeScript, Python, And AI Code Generation: Where Each Fits
A practical architecture for AI-generated systems: TypeScript for contracts and application boundaries, Python for AI and automation, SQL for facts, and CI for independent verification.
AI Agent Identity and Authorization: OAuth for Non-Human Workers
How to design identity, OAuth scopes, policy gateways, audit logs, and least-privilege authorization for AI agents that act across APIs and internal tools.
AI SOC Automation: From Alert Queues To Human-Approved Response Pipelines
How to design AI SOC automation with SIEM normalization, detection-as-code, alert triage, MITRE ATT&CK context, human approval gates, SOAR actions, and incident evidence.
APAC AI Infrastructure: Semiconductors, Data Centers, And Software Supply Chains
Why APAC AI infrastructure is a chip-to-cloud engineering problem: semiconductors, HBM, advanced packaging, data center power, cooling, GPU scheduling, sovereign AI, and software supply chains.