swc swarup.src
  1. 01 lex
  2. 02 parse
  3. 03 opt
  4. 04 link
  5. 05 out
  6. 06 run

$ swc swarup.src --release

Compiling
Swarup E

Lead Engineer at the Centre of Data for Public Good, Bengaluru. Engineer since 2019. I build backend services, Kubernetes platforms, data pipelines and front ends, and I do research on the side.

Scroll to compile

swarup.src UTF-8 · LF
// swarup.src
module swarup_e {
  role  = "Lead Engineer"
  org   = "CDPG"
  city  = "Bengaluru"
  since = 2019
  study = "M.Tech, IISc"

  areas = [backend, devops, data,
           frontend, research, side]

  link [go, kubernetes, flink,
        clickhouse, react, pytorch]

  fn ship(idea) -> system {
    idea |> build |> run |> observe
  }

  export contact
}

01 · lexer

The source, broken into tokens

I am a Lead Engineer at the Centre of Data for Public Good (CDPG) in Bengaluru, and have worked as an engineer since 2019. The work is broad: backend services, DevOps and Kubernetes, front ends, data platforms, research, and side projects.

Since August 2025 I have also been studying for an online M.Tech in Data Science and Business Analytics at IISc.

keywords
5
identifiers
29
strings
4
numbers
3
operators
10
punctuation
26
comments
2

02 · parser

The tree: six areas, one engineer

Backend

  • IUDX Data Sandbox (2022 to 2025). Jupyter notebooks next to datasets. Go API with Keycloak JWT auth and build progress over Server-Sent Events. 70 of 93 backend commits.
  • Data marketplace file service (2025). TypeScript and Express, S3 and MinIO multipart uploads, presigned URLs, Keycloak and policy checks, Redis job queue. 85 of 105 commits.
  • IUDX Video Server (2021). Proxies live camera feeds from city servers behind NAT to the cloud, with load balancing and an API.
  • EpikInDiFi. Listings moved to GraphQL (32% faster), async vendor integrations over RabbitMQ and Socket.IO, MongoDB text search.

DevOps and Kubernetes

  • Hybrid storage and clusters (2023 to now). Ceph, MinIO and Kubernetes across cloud and on-prem, with GPU nodes.
  • JupyterHub and BinderHub on Kubernetes for the Data Sandbox.
  • Multi-kernel notebook images (2026). Python, R, Julia, Scala. CPU image 16.1 GB to 8.1 GB, GPU image 38.9 GB to 26.5 GB.
  • Observability. Prometheus, Grafana and Loki on Kubernetes.

Data

  • IUDX Adaptor Framework (2021 to 2023). Ingest pipeline on Apache Flink, Calcite SQL rules engine over RocksDB state, RabbitMQ and PostgreSQL sinks. Top contributor, 178 of 435 commits. Benchmarked JS engines and moved transforms from Rhino to Nashorn.
  • Analytics at volume (2023 to now). ClickHouse clusters holding billions of rows, sub-second queries, fed by Azure Functions and Event Hubs.

Frontend

  • In-browser Parquet previews via DuckDB in the marketplace file service.
  • Latitude Fintech. A Chrome extension that records clicks into guided product tours.
  • Angular, React, Next.js and ECharts.

Research

  • 4 papers on land use and land cover segmentation from satellite imagery: IEEE InGARSS 2024, two on arXiv in 2024, one on arXiv in 2025.
  • Learning What to Remember (IISc). Streaming memory for vision-language models. The learned memory did not beat the recent-frames baseline, and the report says so.
  • GPT-2 fine-tuned on Mizo (IISc). Test perplexity 680 to 22.5.
  • KYC OCR at EpikInDiFi with Tesseract and Detectron2, 96% accuracy.

Side projects private for now

  • GuideMesh. Record a browser workflow, get a walkthrough that survives code changes. 37 of 38 steps found 50 commits later.
  • videopipe. Videos written as code: HTML and GSAP scenes rendered in headless Chrome.
  • agent-skills. 30 skills for Claude Code and Codex.

03 · optimizer

Smaller, then faster

Optimization is most of the job: the same system, with less to ship and less to wait for. These are the passes I have run on real systems.

notebook image, CPU16.1 → 8.1 GB
notebook image, GPU38.9 → 26.5 GB
listings on GraphQL32% faster
ClickHouse queriessub-second

One multi-kernel image for Python, R, Julia and Scala; listings at EpikInDiFi; analytics over billions of rows.

04 · linker

Linked against the stack

Languages
Go, Java, TypeScript, Python, SQL
Services and auth
Express, FastAPI, Vert.x, GraphQL, Keycloak
Platform
Kubernetes, Docker, Ceph, MinIO, AWS, Azure, JupyterHub, Prometheus, Grafana
Data
Flink, Kafka, RabbitMQ, Event Hubs, ClickHouse, DuckDB, PostgreSQL, MongoDB, Redis
Frontend
Angular, React, Next.js, ECharts
ML
PyTorch

05 · output swarup.bin

Swarup E

Lead Engineer, Centre of Data for Public Good, Bengaluru

  1. Lead Engineer, CDPG
  2. Senior Software Engineer, CDPG
  3. Software Engineer, CDPG
  4. Member of Technical Staff, EpikInDiFi
  5. Intern, Latitude Fintech
  6. Online M.Tech, Data Science and Business Analytics, IISc

papers

  1. An Atmospheric Correction Integrated LULC Segmentation Model for High-Resolution Satellite ImageryIEEE InGARSS 2024
  2. Cross Pseudo Supervision Framework for Sparsely Labelled Geospatial ImagesarXiv 2024
  3. Performance Evaluation of Geospatial Images based on Zarr and TiffarXiv 2024
  4. Comparison of Segmentation Methods in Remote Sensing for Land Use Land CoverarXiv 2025