Case studies

Every industry has a database
that won’t sit still.

Marketplaces, multiplayer games, fleets of trucks, AI products, banks, hospitals, telcos, power grids anywhere data changes, Remac moves it. Four deployments in depth, then a dozen more across the board.

01 · E-commerce02 · Gaming03 · Logistics04 · AI products
Case study 01E-commerceHigh-growth marketplace · 40M SKUs · PostgreSQL → Kafka + Snowflake + Elasticsearch

A 40-million-SKU catalog that's finally consistent everywhere.

One source of truth, fanned out to the warehouse and the search index in real time — no nightly ETL, no 15-minute re-index cron.

01 · The challenge

A nightly job loaded the warehouse and a cron re-indexed Elasticsearch every 15 minutes. Buyers searched for items that existed but weren't indexed yet, and merch dashboards trailed a day behind.

Search stale up to 15 minutesDashboards a day behindFull re-index hammered the DBInventory drift between services
02 · The approach

One Remac binary on the catalog's replication slot fans every change out to three sinks at once — warehouse, search index, and a Kafka topic for downstream services — in order, with delivery guarantees.

PostgreSQL · WALremacSnowflakeElasticsearchKafka
03 · Results
15m → seconds
search freshness
24h → minutes
analytics lag
1 → 3
source → sinks, one config
0
re-index crons left

"New products are searchable within seconds of being added, and the warehouse stays current. We deleted the re-index cron and the ETL DAG in the same sprint."

- Principal Engineer, online marketplace · illustrative
Case study 02GamingLive multiplayer · 12M DAU · multi-region · PostgreSQL → Redis + regional Postgres & MySQL

Leaderboards always current, game state consistent in every region.

CDC streams player data to Redis so leaderboards and session state are never stale, and replicates game tables to Postgres and MySQL nodes across regions — no TTL guesswork, no hand-rolled replicas.

01 · The challenge

Leaderboard data lived in Redis with TTL-based expiry — too short caused miss storms on hot keys, too long served stale standings. Three regions ran a mix of Postgres and MySQL read replicas, all hand-maintained, that regularly drifted, and nobody trusted the data in any of them.

TTL guesswork: stale or thrashingLeaderboards lagged behind playRegions out of syncManual replica wrangling
02 · The approach

Remac streams every row change to Redis so game state is always current, and replicates selected tables to Postgres and MySQL nodes across regions — one binary, one config, heterogeneous replication out of the box.

PostgreSQL · WALremacRedisPostgres · euMySQL · us / ap
03 · Results
CDC-driven
Redis always current
3
regions in sync
1
binary, three sink types
0
TTL tuning or manual replication

"Redis is fed by the WAL now, not by TTLs. Players in every region see the same data, and we stopped babysitting replication scripts."

- Backend Lead, multiplayer studio · illustrative
Case study 03Logistics · MobilityFleet & delivery platform · 250k vehicles · physical + logical · PostgreSQL → S3 + Kafka

Fleet telemetry in the lake, and a recovery story that holds up.

One binary running both modes: continuous data-lake ingestion via logical decoding, and physical PITR via the streaming archiver — replacing fragile scripts and nightly backups.

01 · The challenge

Custom scripts and Airflow DAGs landed vehicle telemetry in S3; nightly backups risked a full day of loss; and services polled each other for the latest vehicle state.

Brittle scripts to land data in S3Nightly backups: up to 24h lossPolling for vehicle stateTwo systems, two failure modes
02 · The approach

Remac streams CDC to S3 as JSONL and to Kafka for downstream services, while the streaming archiver captures raw WAL for point-in-time recovery — all from one process.

PostgreSQL · WALremac · 2 modesS3 · JSONLKafkaS3 · PITR
03 · Results
24h → seconds
recovery RPO
continuous
data-lake ingestion
3 → 0
ingestion scripts
1
binary, both modes

"One binary gives us streaming ingestion to the lake and point-in-time recovery. We retired the backup cron and three ingestion scripts."

- Staff SRE, logistics platform · illustrative
Case study 04AI productsAI product company · catalog RAG · PostgreSQL → embedding → Pinecone · AI Pack

The assistant stopped citing products that don't exist.

Real-time CDC kept a RAG assistant's vector index in lockstep with the catalog — embedding drift fell from hours to seconds, and the embedding cache cut API spend.

01 · The challenge

A nightly cron re-embedded the catalog; between runs the vector DB drifted, so the LLM answered with prices and products that had already changed — and full re-index runs burned the budget.

Hours of embedding driftHallucinated, stale answersExpensive full re-index runsOne more cron to maintain
02 · The approach

Remac + the AI Pack: a row change is captured, the embedding transformer calls OpenAI, and the vector lands in Pinecone within seconds. An embedding cache skips unchanged content.

PostgreSQL · WALremac · embedPinecone
03 · Results
hours → seconds
embedding drift
0
cron jobs
cache
cuts API spend
live
context, always current

This deployment runs on the AI Pack the embedding transformer and vector-DB sinks that bolt onto any paid tier.

Explore the AI Pack →

…and across every other industry.

If it runs on a database, it’s a fit. The same engine, the same single config pointed at whatever your sector needs to move.

Fintech
Real-time fraud analytics

Stream every transaction to the warehouse; fraud runs on live data.

→ batch → real-time
Healthcare
WAL-level audit

Every change captured at the log — complete and cannot be bypassed by application code.

→ 100% completeness
Ad-tech
Event stream to bidders

Impressions and clicks to Kafka in real time, not hourly batches.

→ batch → streaming
EdTech
Progress sync across services

One source of learner state, fanned out to every service in order.

→ N polls → one pipeline
Telco
Network event capture

High-volume CDR and session events streamed straight to the lake.

→ high-throughput streaming
Energy · Utilities
Sensor & meter data

Continuous ingestion of meter reads — no Airflow, no gaps.

→ DAGs decommissioned
Government
Immutable record-keeping

Tamper-evident archive of every change to public records.

→ append-only S3 archive
Travel · Hospitality
Inventory & pricing sync

Rooms, rates, and availability consistent across search and channels.

→ never oversold
B2B SaaS
Customer data fan-out

One change, every downstream service and warehouse, guaranteed.

→ N integrations → 1
Media · Streaming
Recommendation freshness

Feature store updates as soon as a watch or like lands.

→ real-time features
IoT · Manufacturing
Device state sync

Edge and plant-floor changes to the cloud, in order, with retries.

→ ordered, guaranteed
Retail · POS
Omnichannel inventory

Stores, warehouse, and online stock in lockstep, all day.

→ stores + online in sync

Illustrative deployments drawn from common patterns swap in your own logos, quotes, and numbers. Don’t see your industry? If it has a database, Remac fits.

Deploy once. Forget the pipeline exists.

Your data arrives where it needs to be, in order, with delivery guarantees and your team never thinks about it again.

Read the docs