Portfolio · Blog · CV · Book a call · Email · LinkedIn · X
- Roles: Data Engineer, Backend Engineer, or Analytics Engineer (mid-level)
- Experience: 5 years in software since 2021, 3+ of them on production data and backend systems
- Focus: streaming pipelines, lakehouses, GDPR erasure, CI quality gates
- Location: Dhaka, Bangladesh; ready to relocate to Germany, the Netherlands, or Austria, or work remotely across the EU
- Visa: needs sponsorship · Availability: immediate
- Education: B.Sc. Computer Science and Engineering, Daffodil International University, April 2022 to July 2026
- Core stack: Python, Go, Rust, Kafka, Spark, Databricks, dbt, DuckDB, PostgreSQL, Docker, AWS, Azure and GCP
I build streaming pipelines, lakehouses, and the services around them. Most of my work is in Python, Go, and TypeScript, and I like the unglamorous parts: Kafka ingestion, dbt warehouses on DuckDB, GDPR erasure, and schema-contract gates in CI.
Right now I'm building OpencodeHub, a self-hosted Git platform with CI and merge queues, and AegisVision, a multi-camera AI surveillance system. My latest research is Aurora, a modular reasoning architecture with a JOSS paper.
I write about backend systems, data engineering, and building software from scratch on my blog, blog.swadhin.cv.
Languages: English (professional, daily working language) · Bengali (native) · Hindi (conversational) · German (learning).
Email: swadhinbiswas.dev@gmail.com · swadhinbiswas.cse@gmail.com · eu@swadhin.cv
- Co-founded BoringRats and grew it to 1M+ users on a Python, FastAPI, and PostgreSQL backend that served 10K+ requests per minute at under 100 ms. The company was acquired under NDA in 2025.
- Published a JOSS paper and archived DOI for Aurora: GSM8K accuracy rose from 47.2% to 55.6% while confidently wrong answers fell from 19.8% to 2.9%.
- First author on an arXiv paper on noise-robust Bengali dialectal speech recognition.
- I maintain open-source projects with public releases and datasets, including opengrammar at 124 stars and OpencodeHub with 668 tests.
May 2026 to now · Founder and maintainer · OpencodeHub I build and maintain a free self-hosted Git platform with SSH access, CI pipelines, stacked PRs, merge queues, and AI-assisted code review. Go and TypeScript, 668 tests, end-to-end coverage.
2026 to now · Builder · AegisVision A multi-camera AI surveillance and behavioural monitoring system that runs on ordinary hardware. It does real-time face recognition with anti-spoofing liveness checks. Python.
January 2023 to November 2025 · Co-Founder and Lead Data/Backend Engineer · BoringRats (acquired, under NDA) I owned the backend architecture from the first version to 1M+ users: real-time event ingestion, the analytics warehouse, and MLOps workflows. Python, FastAPI, PostgreSQL, Redis, Docker, and AWS. Our REST APIs served 10K+ requests per minute at under 100 ms.
January 2021 to December 2022 · Freelance backend and automation engineer I shipped 8+ full-stack projects for SaaS and e-commerce clients in the US and UK. The Django and PostgreSQL APIs handled 10K+ requests a day, and I cut page load times by 40%.
B.Sc. in Computer Science and Engineering, Daffodil International University, Dhaka. April 2022 to July 2026. Coursework: data structures, algorithms, database systems, machine learning, and software engineering.
- eu-air-traffic: live EU airspace pipeline on Kafka, dbt, and DuckDB. Tracks about 2,000 aircraft with a 15-minute lake, and ships a public Hugging Face dataset and a Zenodo DOI.
- TerraSentinel: $0 anomaly-detection platform over free satellite and sensor data. Four public sources into a versioned lake, dbt and DuckDB build seven gold marts, an IsolationForest scores anomalies, and a Databricks Asset Bundle mirrors the whole pipeline.
- eurostream: GDPR-native streaming lakehouse whose six-layer Article 17 erasure runs in 66.95 ms mean against a 60-second statutory window. Kafka, DuckDB, Turso.
- JustAPI: Python web framework with a Rust core. 766k req/s hello-world, a 0.48 ms p99, and a 12 MB footprint, published on PyPI.
- Aurora: modular reasoning architecture with a JOSS paper. GSM8K accuracy 47.2% to 55.6%, confident errors down to 2.9%.
- opengrammar: privacy-first Grammarly alternative with a 156k-word offline engine, five runtimes, and 124 stars.
- OpenCodeHub: self-hosted Git platform with CI, stacked PRs, merge queues, and AI review. 668 tests.
- AegisVision: multi-camera AI surveillance and behavioural monitoring in Python, with real-time detection.
I write about backend systems, data engineering, and building software from scratch on my blog, blog.swadhin.cv.
- Designing data pipelines that stay cheap · 24 Sep 2026
- Designing an API for 100,000 requests a second · 24 Sep 2026
- Flare: self-hosted webmail for my own domain · 21 Sep 2026
- A filesystem event log in SQLite · 20 Sep 2026
More at blog.swadhin.cv · RSS · Atom
- A Unified Denoising and Adaptation Framework for Self-Supervised Bengali Dialectal ASR. First author, with Imran and Tuhin Sheikh. arXiv:2509.00988 · PDF WavLM-based framework with two-stage fine-tuning for noise-robust dialectal recognition. Tested across dialects from clean audio to low signal-to-noise ratios, it beats fine-tuned wav2vec 2.0 and multilingual Whisper.
- Aurora (ART): A Modular Reasoning Architecture for Reliable and Efficient Neural Systems. Lead author. paper · DOI Six-stage transformer with an explicit uncertainty channel: GSM8K 47.2% → 55.6%, confidently wrong answers 19.8% → 2.9%.
- An Empirical Benchmark Dataset for Paillier-Based Privacy-Preserving REST API Gateways. Zenodo DOI Per-request telemetry benchmark for homomorphic encryption overhead in REST API gateways.
Full list with abstracts: swadhin.cv/research
Core: Python · Go · Rust · TypeScript · SQL · Kafka · Spark · Databricks · Airflow · dbt · DuckDB · PostgreSQL · Redis · FastAPI · Django · Docker · AWS (Glue, Redshift, Athena) · Azure (Data Factory) · GCP (Dataflow, Dataproc) · PyTorch
Working knowledge: ClickHouse · Kubernetes · Terraform · Airbyte · Dagster · Prometheus · LangChain · OpenAI / local LLM APIs
Data engineering · Cloud · Databases · AI · Backend · full stack
DATA ENGINEERING RESEARCH eu-air-traffic live EU airspace pipeline contexa versioned llm agent memory TerraSentinel databricks satellite anomaly DOOMSDAYCS offline cs encyclopedia eurostream gdpr-native streaming lakehouse FAANG-Playbook 1,400+ leetcode problems DEVOPS TOOLS OpencodeHub git platform w/ ci pipelines veet universal app uninstaller gvx the pnpm of python lsf ls with nerd-font icons HiFiLinux audiophile audio for linux fetchx neofetch rewritten in rust Ghost free & open coding tool BACKEND ZenDownload download anything, one place JustAPI zero-copy rust web framework vscode-android a real ide for android warren rootless cli runtime MACHINE-LEARNING Aurora modular reasoning architecture OTHERS AegisVision multi-camera ai surveillance linuxy one-click appimage runner Ecoguard self-hosted llm inference gw de-omarchy modern desktop, no omarchy opengrammar open-source grammarly alt Mervelas ai coding cli built on bun VidoLib lag-free media engine moonshell personal qml linux rice
Case studies for the featured work: swadhin.cv/projects
- Portfolio: swadhin.cv
- Blog: blog.swadhin.cv · RSS
- CV (printable): swadhin.cv/cv
- Book a call: cal.com/swadhinbiswas
- LinkedIn: @swadhinbiswas
- Email: swadhinbiswas.dev@gmail.com · swadhinbiswas.cse@gmail.com · eu@swadhin.cv
- X: @swadhin_sh
- YouTube: @BitsWar
off the clock: street-food hunter · anime marathoner · Mystery & mythology story hunter