Douglas EduardoSenior Backend Engineer · Brazil (UTC-4)

Backend systems that stay correct under load.

Go, TypeScript, and PostgreSQL, with six years on multi-tenant SaaS, high-throughput APIs, event-driven systems, and a lot of time spent in query plans.

Remote · US/EU hours overlap · Contractor

Now

Founding engineer at Solaris Neuro, owning the backend, data layer, and infrastructure of a multi-tenant clinical platform for ABA therapy clinics.

Six requests race for the same room and hour. Switch the guard and watch it stop being possible.

Guard the write with

Isolation level: read committed — and it still races

2 rows committed for one room

Two transactions read the same empty slot before either of them wrote. Both inserts succeeded and no error was raised, so the clinic finds out from the patient.

SELECT 1 FROM booking WHERE room = 3 AND slot && $1;
-- no row found, so:
INSERT INTO booking (room, slot) VALUES (3, $1);
Evidencethree kinds of claim

What these systems did in production

Eliminated

occurrences since · by construction
  • Double-booked rooms and therapists

    Exclusion constraints over tstzrange, replacing application-level retries · Solaris Neuro

  • Cross-tenant reads

    Row-level security with transactional tenant context, verified across 40K+ concurrent queries · Solaris Neuro

  • Duplicate charges on provider retries

    Idempotency keys, webhook signature verification, transactional reconciliation · Solaris Neuro

  • Duplicate sends across worker restarts

    Composite uniqueness constraints and atomic claim operations · Project Broadcast

  • Oversubscribed reservation capacity

    Row-level locking and database constraints under READ COMMITTED · Coopers Digital

Sustained

steady state, not peak
10K+
msg / min
outbound throughput, carrier 429s under 0.5%
100K+
events / day
idempotent consumers with dead-letter queues
1,000+
connections
concurrent WebSockets on 2 vCPU
50M+
rows
monthly-partitioned event table, queries stable

Reduced

each pair scaled to its own before

Clinical dashboard

p95
1.8s220ms-88%

Cut N+1 aggregation, added partial indexes and LATERAL joins, moved reporting to incrementally refreshed materialized views

Booking write

p99
310ms48ms-85%

Let PostgreSQL reject conflicts instead of retrying them in the application

Scheduled dispatch

p99
47sunder 5s-89%

Batch claiming with FOR UPDATE SKIP LOCKED and lease expiration across workers

Delivery-status lag

typical
~90sunder 3s-97%

Decoupled webhook ingestion from asynchronous processing

Folder listing

large accounts
1.4s~100ms-93%

Materialized-path hierarchy queried with recursive CTEs

Work2020 — present

Four roles, one specialty

2025 —

Remote

Solaris Neuro

Founding Engineer

Multi-tenant clinical platform for ABA therapy clinics. Own backend architecture, data layer, and infrastructure.

  • Architected PostgreSQL row-level security for shared-schema multi-tenancy, propagating tenant context transactionally through PgBouncer and validating isolation with a concurrent test harness covering 40K+ queries with zero cross-tenant reads.
  • Eliminated double-booking across therapists, rooms, and patients by replacing application-level retries with PostgreSQL exclusion constraints over tstzrange, reducing booking conflicts from recurring production incidents to zero while cutting p99 write latency from 310ms to 48ms.
  • Reduced clinical dashboard p95 latency from 1.8s to 220ms by eliminating N+1 aggregation, introducing partial indexes and LATERAL joins, and moving expensive reporting workloads to incrementally refreshed materialized views.
Show 4 more
  • Designed an event-driven processing layer with RabbitMQ, idempotent consumers, dead-letter queues, and bounded retries, processing 100K+ events/day while preventing poison messages from blocking unrelated workloads.
  • Built a Go WebSocket gateway for live therapy sessions with bounded write buffers and backpressure-aware connection handling, sustaining 1,000+ concurrent connections on 2 vCPU while using substantially less memory than the previous Node.js implementation.
  • Partitioned high-growth clinical and audit tables by month with automated retention and BRIN indexes, keeping time-based queries stable as the primary event table exceeded 50M rows.
  • Implemented payment processing with webhook signature verification, idempotency keys, replay protection, and transactional reconciliation, eliminating duplicate-charge incidents caused by provider retries.

2023 - 24

Fort Worth, TX · Remote

Project Broadcast

Software Developer

SMS and voice marketing platform serving 30K+ monthly active users.

  • Rebuilt outbound message processing around BullMQ and Redis token-bucket rate limiting, increasing sustained throughput to 10K+ messages/minute while reducing carrier-side 429 responses from 4% to below 0.5%.
  • Decoupled webhook ingestion from asynchronous processing, reducing inbound delivery-status lag from approximately 90 seconds to under 3 seconds during high-volume customer broadcasts.
  • Made broadcast delivery idempotent across worker restarts using composite uniqueness constraints and atomic claim operations, eliminating duplicate sends caused by select-then-insert race conditions.
Show 3 more
  • Reworked scheduled broadcast dispatch using FOR UPDATE SKIP LOCKED batch claiming across multiple workers with lease expiration, reducing p99 dispatch latency from 47 seconds to under 5 seconds.
  • Designed a materialized-path folder system in PostgreSQL and replaced application-level traversal with recursive CTEs, reducing folder-listing latency from 1.4s to approximately 100ms for large customer accounts.
  • Introduced structured logging, request tracing, and correlation IDs across backend services, reducing average production incident investigation time from roughly 45 minutes to under 10 minutes.

2022 - 23

Remote · Brazil

Coopers Digital

Software Developer

Engineering across reservation, commerce, and third-party integration systems.

  • Designed the concurrency model for reservation inventory after READ COMMITTED transactions allowed competing requests to oversubscribe capacity, using row-level locking and database constraints to eliminate recurring overbooking incidents.
  • Built an idempotent RabbitMQ order-ingestion pipeline keyed by external order IDs, processing 200K+ orders/month across dozens of merchant integrations without duplicate order creation.
  • Reduced peak database load by approximately 60% by introducing Redis caching with request coalescing and TTL jitter, preventing cache stampedes on high-traffic catalog keys.
Show 2 more
  • Optimized merchant reporting queries using window functions, composite indexes, and execution-plan analysis, reducing p95 dashboard latency from several seconds to below one second.
  • Automated Brazilian fiscal-document workflows and retry handling against external government services, removing recurring manual operational work from merchant onboarding and order processing.

2020 - 22

Remote · Brazil

SoloEd

Mid-Level Software Developer

Full-stack product development for GoMc, an internal social platform used across a 30K+ employee organization.

  • Built real-time voting and notification services using Node.js, PostgreSQL, Redis, and WebSockets, maintaining consistent vote state across thousands of concurrent connections during campaign peaks.
  • Implemented Elasticsearch-backed search with fuzzy matching and abbreviation expansion, handling tens of thousands of daily queries while substantially improving successful results for typo-heavy searches.
  • Replaced offset pagination with cursor-based pagination and moved image processing off the request path, reducing feed response times by approximately 40% as the underlying dataset grew past several million records.
Show 1 more
  • Delivered responsive web experiences, email campaigns, and landing pages for enterprise clients including Bosch.
Benchmarksthree labs

Built to find out, not to ship

Multi-Tenant Isolation Benchmark

  • Go
  • PostgreSQL

Built a reproducible benchmark comparing shared-schema PostgreSQL RLS, schema-per-tenant, and database-per-tenant architectures. Measured query latency, connection-pool pressure, migration complexity, and operational overhead across increasing tenant counts.

WebSocket Gateway

  • Go
  • Redis
  • Prometheus

Built a horizontally scalable WebSocket gateway with bounded write buffers, slow-consumer eviction, Redis-based cross-node routing, connection lifecycle metrics, and load tests targeting high connection counts and message fan-out.

Distributed Job Processing Lab

  • Go
  • PostgreSQL
  • Redis
  • RabbitMQ

Implemented competing job-processing strategies using database locking, Redis queues, and RabbitMQ. Benchmarked throughput, retry behavior, duplicate execution, worker crashes, backpressure, and recovery under contention.

Stackfive groups

What I reach for

Languages
GoTypeScript/JavaScriptPythonSQL
Backend
Node.jsNestJSRESTGraphQLWebSocketsgRPC
Data
PostgreSQLRedisElasticsearchSQLite
Distributed Systems
RabbitMQKafkaBullMQidempotencyrate limitingtransactional outboxdead-letter queuesconcurrency control
Infrastructure
DockerKubernetesDocker SwarmAWSGitHub ActionsPrometheusGrafanaLinux

EducationAssociate Degree in Systems Analysis and DevelopmentFederal Institute of Mato Grosso do Sul (IFMS)2020-2022

Contact

Available now for senior backend work

Remote from Brazil with US and EU hours overlap, as a contractor or through an EOR. Send the problem you are stuck on — schema, throughput, or a race you cannot reproduce — and I will tell you what I would try.