< I build the data infrastructure behind a Swedish B2B sales intelligence product β pipelines processing 10M+ records daily into a 300M+ document Elasticsearch index, an S3 + Apache Iceberg lakehouse queried through Athena, and search clusters I've re-architected to cut AWS costs 43%. Since 2023 I've been building the AI layer on top: a self-hosted LLM on llama.cpp, MCP servers, LangChain based agents, and a RAG proof-of-concept over our Elasticsearch data that lets clients query companies, news, and recruitment in natural language. Python Β· SQL Β· Elasticsearch/OpenSearch Β· AWS Β· Iceberg Β· Athena Β· Glue Β· Scrapy Β· Crawl4ai Β· Playwright Β· FastAPI Β· llama.cpp Β· LangChain Β· MCP Β· RAG />
Insights and experiences from my journey in data engineering and software development.
SQL window functions are powerful tools for analytical queries, especially in data-heavy environments. They were created to address limitations in traditional SQL operations, particularly for performing analytical calculations across related rows without collapsing the result set.
It is a very simple SQL concept but heavily used. Usually, we use it with one joining / matching condition, but sometimes there can be multiple conditions depending on the scenar
We need to use field data caching to utilize fast aggregation and sort queries. However, caching too many fields can degrade Elasticsearch performance by reducing available JVM memory.
Do you know why the π£πΏπΌππΆππΆπΌπ»π²π± π§π΅πΏπΌππ΄π΅π½ππ (π πΆπ/π) and ππ’π£π¦ (Input/Output Operations Per Second) are very important stuffs to consider based on requirements?
We may encounter an yellow cluster status of an AWS Elasticsearch cluster. Do we know why dpes the cluster run in yellow state?
-lifecycle-elasticsearch.webp&w=1080&q=75)
An Entire Lifecycle of An Indexing Request (Single & Bulk) in Elasticsearch. When we send a document to be indexed, here is the sequence of events that must happen before our client/service gets a 200 OK or 201 Created response.
Trusted by Industry Leaders
Chief Technology Officer at Goava
September 11, 2026, HΓ₯kan managed Md. Mehadi directly
I have had the pleasure of working with Mehadi Hasan at Goava Sales Intelligence AB since December 2020. He was promoted to Senior Software Engineer in March 2026.
Mehadi is a capable and dependable engineer with experience in backend development, data pipelines and cloud infrastructure. He takes responsibility for his systems from design through to production and pays attention to detail.
His work at Goava includes ownership of our primary company data pipeline, development of the Twingly API integration and news pipeline powering our company news and recruitment signals, and the design of our contacts data pipeline. The latter consolidates director and people data from multiple sources with reliable deduplication and full change history.
Mehadi has also made important contributions to maintaining and improving our OpenSearch infrastructure. He has worked through production incidents with a focus on identifying and addressing the underlying causes. His redesign of our sharding strategy and reindexing of two of our largest indices enabled us to right-size the cluster and commit to reserved instances. We expect these changes to reduce our AWS OpenSearch spend by around 40%.
Mehadi is proactive and takes ownership of his work. He identifies problems and cost inefficiencies and proposes practical solutions. He considers security in his design decisions and is open about trade-offs, risks and mistakes. I appreciate his honesty and straightforward communication.
I can confidently recommend Mehadi for backend development, data engineering or cloud infrastructure work. He would be a valuable addition to any development team.
