01 / Data correctness
Make reruns part of the design
At Mobobeat, I used database constraints to keep repeated ETL runs from duplicating records, with retries and an email alert on final failure.
Ten years of backend engineering, most of it in Go. Ad exchanges bidding in milliseconds, carrier messages turning into delivery statuses, hospital imaging routed to AI vendors. Different industries, same problem underneath — make two systems that know nothing about each other behave like one.
Industries
Mobobeat · 2016–2021
Agathsya · 2021–2022
Kramp · 2022–2025
Founda Health · Joined 2025
01 — About
I'm Hemraj, a backend engineer based in Almere, in the Netherlands. Over the past decade, I've built software for advertising, logistics, e-commerce and healthcare.
At Mobobeat in Goa, my work covered ad-serving APIs, real-time bidding and data pipelines. At Agathsya, I led the team building a truck aggregation MVP. At Kramp in the Netherlands, I worked on shipping integrations and the services behind product lifecycle, notifications and visibility.
At Founda Health, that integration work extended to hospital imaging and healthcare standards. I built core services for AI Hub and took the reworked architecture through to its final demo.
The connection across these roles is practical: turning external data into something another system can use. That has meant translating carrier messages, processing product feeds, moving analytics data and coordinating imaging workflows.
01 / Data correctness
At Mobobeat, I used database constraints to keep repeated ETL runs from duplicating records, with retries and an email alert on final failure.
02 / Integration
At Kramp, I turned carrier messages into delivery statuses and built a connector that transformed product XML for downstream services.
03 / Observability
At Founda, I set up OpenTelemetry pipelines into ClickHouse and wrote scripts to backfill historical telemetry with progress checks for reruns.
02 — Focus
Go services for product lifecycle, shipping, notifications and logistics. My experience includes domain modelling, event-driven communication, DDD and CQRS, with GCP and Terraform used to provision services at Kramp.
Integration work in the Dutch healthcare landscape: DICOMweb for imaging, FHIR R4B for patient data and vendor notifications, and mCSD for care services discovery. At Founda, I brought these pieces together with Temporal workflows.
A Kotlin service rewritten in Go for a 50%+ throughput gain. An ad-serving backend taken from 12k to 20k requests per second while infrastructure cost fell about 40%. These projects combined backend implementation with attention to throughput, latency and infrastructure cost.
03 — Experience
Joined Nov 2025
Amsterdam, NL
Healthcare interoperability for the Dutch market. I reworked the architecture of AI Hub — the platform that routes hospital imaging studies to third-party AI vendors — and led it through to the final demo.
May 2022 — Oct 2025
Utrecht, NL
Three and a half years on a Go microservices platform behind agricultural parts distribution across Europe. Shipping, product lifecycle, translations, notifications — mostly the services other teams end up depending on.
Apr 2021 — Feb 2022
Bengaluru, IN
Led the team that shipped the MVP of a truck aggregation platform for the Dubai market — matching consignors and consignees into trips, modelled as a graph because the relationships were the product.
May 2016 — Mar 2021
Goa, IN
Five years in adtech, which is a good place to learn latency. I built the traffic backend and then the exchange that bid on it.
Education St. Xavier's College — BSc Computer Science, 2011–2015 · Goa, IN
04 — Selected work
A closer look at three projects: the systems involved, what I built, and the mechanisms behind the implementation.
01 / Founda Health
AI Hub brings imaging transfer, patient data and vendor communication into a coordinated workflow. I reworked its architecture and built the core Go services.
A hospital imaging study needs to reach a third-party AI vendor with the relevant patient context. Image transfer, data enrichment and notifications involve different interfaces and need to be coordinated.
I built a DICOMweb streaming service in Go, Temporal workflows for orchestration, and FHIR R4B integrations for vendor notifications and patient data enrichment.
DICOMweb handled imaging transfer, FHIR provided the patient-data and notification interfaces, and Temporal coordinated the workflow steps. This gave the integration an explicit workflow across the participating services.
I led the reworked architecture through to the final demo. This account describes the implementation and integration work completed for that milestone.
02 / Mobobeat
I built scheduled Scala/Spark jobs to move large ad-click datasets from Cassandra into MariaDB, with safeguards for reruns and failed jobs.
The pipeline extracted ad-click data from Cassandra, transformed it with Scala and Spark, and loaded it into MariaDB on a schedule.
I used composite unique constraints in MariaDB to prevent duplicate records when data was processed again. The destination database enforced uniqueness rather than relying only on the job's execution history.
Jobs retried three times and sent an email after final failure. Retries covered recoverable failures; the alert made an exhausted retry visible for follow-up.
I also built scheduled advertiser API integrations to create, update and delete campaigns, and automation that disabled active campaigns with low install conversion.
03 / Kramp
I built a STEP-to-Offer connector that ingested product XML, transformed it, stored a PostgreSQL copy for deduplication and published downstream.
The connector sat between the STEP product-data source and the downstream Offer system. Its job was to translate the incoming XML into the form needed by the next part of the platform.
I implemented ingestion and transformation, stored a copy in PostgreSQL for deduplication, and published the transformed data downstream.
The stored copy supported duplicate detection during processing. Transformation and deduplication were part of the connector's responsibility at the integration boundary.
My work also includes an OpenRTB ad exchange and publisher SDK, a truck aggregation MVP, product lifecycle services, and a Kotlin-to-Go service rewrite. See the career timeline
05 — Activity
06 — Stack
Go is my main backend language. Scala and Python also feature in my data-pipeline work.
GCP and Terraform for service infrastructure at Kramp; containers and deployment tooling across projects.
Operational data, analytics and ETL, including PostgreSQL deduplication and Cassandra-to-MariaDB pipelines.
Domain-oriented services at Kramp and Agathsya, workflow orchestration and telemetry at Founda.
Imaging transfer, patient data, vendor notifications and care services discovery.
Led the logistics MVP team at Agathsya and the reworked AI Hub architecture through its final demo at Founda.
07 — Contact
Interested in Go, distributed systems and healthcare interoperability — especially anything in the Dutch or wider European interop landscape. Also happy to talk adtech latency tricks with anyone who will still listen.