I'm Rajeev — a data engineering team lead in Kathmandu. I spend most of my weekdays building production data pipelines for blockchain analytics and social intelligence, and most of my weekends running, watching F1, or arguing about midfield football.
I lead a small data team at 108 Capital. We ship pipelines that index multi-TB blockchain history, scrape the strange corners of social media, and turn audio into searchable text — the kind of plumbing that's unglamorous until it isn't.
I care about systems that don't fail silently, code reviews that improve taste, and Saturdays that smell like wet pavement after a long run. I studied computer engineering at Pulchowk Campus in Nepal and have been writing data infra full-time since 2023.
This site is mostly for me. A spot to keep notes, list a few things I'm proud of, and remember that there's life outside the warehouse.
Benchmarking across cluster shapes for our analytics workloads.
Third pass. Different highlighter color this time.
Long runs at 5:30 a.m. before the city wakes up.
The new regs are weirder than expected. Loving it.
Lead a team of engineers shipping production data infrastructure. Owned architecture, deployment, and monitoring across five products in our first year together — from on-chain indexers to NLP-on-warehouse sentiment pipelines.
Built the on-chain indexer from scratch as sole owner — a ClickHouse warehouse with dbt + Prefect ETL keeping EVM data under one-minute latency at multi-TB scale. Then built the Twitter pipeline and started the Spark migration.
Built a multiplayer mobile Ludo game in Flutter and owned the real-time communication layer over WebRTC. Different stack, same taste for low-latency systems.
Production systems shipped at 108 Capital. Each one has a stack story and an outage story — happy to tell either over coffee.
Multi-chain blockchain data platform covering Ethereum Mainnet, Base, BSC, Arbitrum, and Optimism. Sub-one-minute end-to-end latency using Kafka stream processing and ClickHouse UDFs for real-time ABI decoding.
Twitter pipeline covering account management and ETL orchestration with Airflow + ClickHouse. GPU-enabled NLP UDFs run on a distributed cluster for in-warehouse sentiment analysis and entity extraction.
Media transcription pipeline that tracks new YouTube channels and podcast feeds, downloads content, and transcribes with AI. Orchestrated with Plomberry; Celery handles the async background work so the queue never blocks the front end.
Multi-platform newsletter and article scraper covering Substack and Seeking Alpha, including paywalled content via authenticated browser sessions. Ingested into SurrealDB through Browserless workers.
Subreddit and user-tracking ingestion pipeline. SurrealDB doubles as both data lake and warehouse, removing the need for a separate ETL layer — fewer moving parts, fewer places to fail silently.
Designing a benchmarking framework across cluster configurations and sizes. Documenting the findings as we go — the answer is “it depends”, and the interesting work is figuring out on what.
A non-exhaustive list of obsessions that crowd out engineering on the weekend. None of them are productive. All of them are the point.
Started after my wrist gave up on too much typing. Now somewhere between "stubborn jogger" and "slow marathoner". Targeting sub-4 in October.
Race day starts before lights out. I keep an embarrassingly detailed spreadsheet of pit-stop deltas — old habits.
The bikes lean past 60° and pretend gravity is optional.
I watch midfielders like other people watch the ball. La Liga and the Premier League weekends are sacred.
Systems books, distributed-systems papers, the occasional novel when my brain needs a hard reset.
Kathmandu's hills are the best debugging environment I've found. One hour out, one bug fewer.
Engineering postmortems, stray opinions, the occasional running journal. New entries at the top.
All notesI'm slow on Twitter, fast on email. Most replies within a day.