Surajit Das
Surajit Das

Hi, I am Surajit.

I build data pipelines and machine learning systems for things that move: vehicles, fuel and money. I am a senior backend and data engineer based in Nairobi, Kenya, and I have been writing production Python since 2017.

On the side, I:

  • run my own vehicle fleet, which gives me messy, real data to work with
  • train ML models on that data, to learn by doing
  • write about what I build on Medium
  • build small AI agents, carefully

Away from the keyboard, I have trained boxing and kickboxing for about 18 years. Most days start at 5 AM with pad work.

Email GitHub Medium

Writing

All writing on Medium

Selected projects

  • Fleet telemetry pipeline

    My own ingestion service for GPS trackers. Binary packets over TCP, duplicates dropped, queued through Redis, live state in Postgres and full history in ClickHouse.

    Python, Redis, PostgreSQL, ClickHouse

  • Fuel theft detection

    Finds fuel siphoning in multi tank diesel trucks while ignoring slopes, refuels and fuel moving between tanks. Rules first, tested on labelled synthetic data.

    Python, pandas, signal smoothing

  • Siphyy

    Open source framework that maps any telematics provider into one schema, runs cheap detectors, and sends only the interesting events to an LLM.

    GitHub, live demo, on PyPI

  • Driver risk and battery models

    My first ML models, trained on my own tracker data. A decision tree on battery voltage flagged a car that looked fine, and we pulled its battery to check.

    pandas, scikit-learn, Jupyter

  • Trickstar

    A browser agent that finds sign up pages with a vision model, verifies every click, and replays routes that worked without calling the model again.

    Python, Playwright, Qwen3.5 9B, SQLite

  • Pen

    An email outreach assistant I built for myself. It finds prospects every night, drafts one email per person, and sends only after I approve.

    Python, Gmail API, scheduled runs

Earlier work

  • 2024 to 2025Senior engineer, fintech. Ledger ETL and a reporting service on Kafka, Redis and partitioned Postgres.70% faster reports, 5× throughput
  • 2022 to 2024Senior backend engineer, payments, remote. Payment services with retries, idempotency and audit logs.99.9% of events captured, payments 40% more reliable
  • 2020 to 2022Senior backend engineer, ERP and supply chain. Kafka event pipelines and payment integrations.10× throughput, 80% lower latency
  • 2018 to 2020Backend engineer, healthcare. Billing, M-Pesa collections and reconciliation.80% of reconciliation automated, claims 60% faster
  • 2017 to 2018Backend engineer, payments. Mobile money gateway and transaction processing.10K+ transactions a day

How I use AI

I got into software because I love to code. I still do, so I am picky about where AI goes.

Where it helps me

  • Reviewing my pull requests before anyone else sees them.
  • Thinking through design, specs and tradeoffs out loud.
  • The fuzzy part of a problem, after cheap rules have filtered out the rest.

Where I keep it out

  • Writing my code. I went back to writing it by hand. It is slower, and I know exactly what I ship.
  • Anywhere a simple rule already gives the right answer.
  • Anything I cannot verify.

Now

  • Growing a small FastAPI service into a full data cycle: ingestion, a proper pipeline, a few ML models and a little AI on top.
  • Learning Docker, Terraform, Kubernetes and AWS, a little every day.
  • Reading about change data capture for ETL, and learning Rust slowly.

Contact

If you are working on data heavy problems, especially anything that moves, I would like to hear about it. Email is the best way to reach me: surajit.das.03211@gmail.com

GitHub Medium Stack Overflow