Rohit Swami
India Resume ↗

Rohit Swami engineer and product builder India

Systems that hold up at 3am.

I'm the engineer you hand the thing that's on fire. The pipeline that runs all night, the graph too big to render, the service that dies under real traffic. Five and a half years of turning those into things that are fast, boring and still running. Fifty-odd of them now in production.

Fig. 1 Not a real trace: the pipeline, the graph and the service from above, profiled before and after. Hover a frame; click one to fix it.

on fire

§1

Engineer, mentor, and hands-on architect

Portrait of Rohit Swami
Fig. 2 The author, at one bit per pixel. Hover to develop.

I build the unglamorous machinery real products run on. Across 5.5+ years and 50+ production projects the pattern is usually the same: something slow, fragile or expensive comes in, and something fast, boring and cheap goes out.

What I build
  • Low-latency microservices and APIs that hold under real traffic
  • Data pipelines that survive production volume, not just the demo
  • Real-time infrastructure on Redis and WebSocket
  • GenAI systems with LLMs and RAG, shipped rather than prototyped
How I work
  • Profile first, then optimise. Guessing is expensive
  • Own it end to end: infrastructure, CI/CD, and the 3am page
  • Test automation and code review as defaults, not afterthoughts
  • The security work nobody volunteers for, including HIPAA and GDPR
Where I'm headed
  • A hands-on tech lead role, owning architecture without leaving the code
  • Teams that ship, measure, and fix what the numbers show
  • More mentoring. It is the part I would keep above everything else

Technical skills

Languages & scripting
Python · TypeScript · JavaScript (ES6+) · SQL · R · Bash · Go · HTML/CSS · Data Structures & Algorithms
Cloud (AWS)
EC2 · AWS Batch · Lambda · EKS · S3 · RDS / Aurora · DynamoDB · API Gateway · Step Functions · SQS / SNS · CloudWatch · VPC
DevOps & infrastructure
Docker · Kubernetes · Terraform · AWS CDK · CloudFormation · GitHub Actions · CI/CD
Data engineering
Polars · Pandas · Parquet · dbt · SQLAlchemy · PostgreSQL · MongoDB · MySQL · Redis
Frameworks & frontend
FastAPI · Flask · Django · React · Next.js · Node.js · Express.js · Redux · R Shiny · Streamlit · WebSocket · GraphQL
AI & GenAI
LLMs · RAG · Prompt engineering · FAISS · Qdrant · Vector databases
Testing & ways of working
PyTest · Jest · Cypress · Team leadership · Architecture design · Mentoring · Agile · Code review · HIPAA / GDPR

If that is the kind of engineer you are hiring, let's talk.

§2

Where I've worked

From analysing public sentiment for the Government of India to architecting real-time microservices that process 100K+ biomedical entities per run.

work
MyGov upGrad InterviewBit Ignite Elucidata
community
MSTC lead
study
B.Tech, LPU
Udacity
papers
projects
products

 

Fig. 3 Ten years, drawn as a flame chart. Hover to scrub through time; click anything to jump to it.

  1. Feb 2022 → now

    4 yrs 8 mos

    Senior Software Engineer

    Elucidata · New Delhi (remote)

    • Rebuilt the graph engine from R to Python/Polars for 10× faster runtime, 80% less memory, scaled to 100K+ entities across 1,000+ pathways.
    • Optimised KEGG pathway rendering: runtime 22 minutes → under 1 minute, UI latency 15 minutes → under 1 second.
    • Built a production Streamlit + Python app with bi-directional S3↔EC2 sync and Elastic IP deployment, giving instant h5ad visualisation, runtime from hours to seconds.
    • Developed a React + Next.js + FastAPI platform for OMOP table mapping with PostgreSQL validation, cutting manual mapping effort by 70%.
    • Designed AWS Batch pipelines running 10,000+ jobs/month on a spot/on-demand mix, cutting infrastructure costs by 28%.
    • Architected a Redis + WebSocket microservice (EC2, VPC, CI/CD) delivering sub-100ms notifications at scale.
    • Built production Shiny apps and React/Node.js/Python POCs that powered client demos and secured new contracts.
    • Shipped production GenAI apps using LLMs, RAG and vector databases (FAISS / Qdrant) for real-time semantic search and summarisation on domain-specific data.
    • Developed 10+ interactive dashboards in R Shiny and React, including a Yale University data explorer later published in a peer-reviewed journal.
    • Diagnosed and fixed a major memory leak in tooltip rendering for 90% less memory and no more Shiny disconnects.
    • Mentored three engineers, introduced CI/CD automation (85% test coverage) and code review, cutting 40% fewer production issues.
    • Hardened infrastructure security: HIPAA/GDPR-compliant handling across S3, RDS and VPC, passing a third-party audit with zero critical findings.

    Python · Polars · FastAPI · React · Next.js · AWS Batch · Redis · WebSocket · R Shiny · LLMs / RAG

  2. Jan 2021 → Feb 2022

    1 yr 2 mos

    Software Engineer

    Ignite Solutions · Pune (remote)

    • Built demand-forecasting pipelines (Python, Airflow) reducing MAPE by 15% for 500K retail energy meters.
    • Modelled 100M credit-card transactions, boosting upsell precision to 90% in the pilot region.
    • Created a Flask-based recruitment portal integrated with Google Workspace, cutting hiring cycle time by 30%.

    Python · Airflow · Flask · Forecasting

  3. Aug 2019 → May 2020

    10 mos

    Software Engineer, Intern

    InterviewBit · Bengaluru, India

    • Implemented an end-to-end auto-suggestion tool measuring semantic similarity between student queries and existing solutions, deployed as a REST API on AWS EC2 with a Flask backend, reducing query resolution time for 72% of students from 2 days to 13.3 minutes on average. Built a KPI dashboard on top using SQL.
    • Built an NLP tool to calculate the percentage of code vs. text in TA responses, increasing overall student–TA interaction.
    • Wrote scripts to email and migrate 1,500+ students from Flock to the open-source platform Mattermost.
    • Revamped InterviewBit's web pages and created content for Data Science, Machine Learning and Deep Learning tracks.

    Python · NLP · Flask · AWS EC2 · SQL

  4. Jun 2019 → Jul 2019

    2 mos

    Data Science, Intern

    upGrad · Remote

    • Mentored 200+ students, providing clear, positive, line-by-line actionable feedback on their submitted projects using upGrad's code review tool for data science courses.

    Python · Code review · Mentoring

  5. Jun 2018 → Jul 2018

    2 mos

    Data Analyst, Intern

    MyGov (MoEIT, India) · New Delhi (remote)

    • Analysed public sentiment on government policies and campaigns across social media.
    • Exported over 4,000 tweets with the Twitter API and built a hybrid solution classifying each tweet as positive or negative with a KNN algorithm.
    • Performed data mining operations on websites for various internal purposes.

    Python · Twitter API · Sentiment analysis

  6. Aug 2017 → Sep 2018

    1 yr 2 mos

    Lead Organizer

    Microsoft Student Technical Community, LPU · India

    • MSTC hosts community events guiding professionals across different technologies. It is a platform where like-minded people come together to share and learn about technology.
    • As a lead speaker, shared knowledge on front-end technologies.

    Community · Public speaking · Front-end

§3

Products I've built

Six of my own, from 2026: the apps, their websites and the backends behind them. The lights are live. Your browser is checking each site right now.

How four of them work, written up: parsing chats for YAPD, ClearOwe's sealed sync, CitizenDays' day counting and the exam engine.

§4

Things I've built

A selection of the public ones: personal, open-source and hackathon work. Most of the 50+ I've shipped live behind company walls.

Nameplate

Oct 2026 · TypeScript · VS Code, Cursor, Windsurf and other editors built on VS Code

A free VS Code extension that puts the project's name in the status bar and gives every window a color of its own, so a row of identical windows stops being identical.

Four VS Code windows that look identical, then the same four with Nameplate: each shows its project name in a status bar of its own color.
Fig. 4 Four projects without Nameplate, then with it. Every name is found and every color chosen without a single setting.

Problem statement: with several projects open, every VS Code window looks the same, which is how a command runs in the wrong terminal or a commit goes to the wrong repository.

  • Finds a name without asking: the folder unless it's generic (app, src, frontend), then manifests such as package.json, Expo's app.json or pyproject.toml, then the Git remote, keeping any deliberate capitals.
  • Derives the color from the Git remote, so a repository looks the same on every machine, from eleven colors designed in OKLCH to look unlike each other, with text that always clears WCAG AA.
  • Keeps open windows apart: colors are compared by how they look, as OKLab distance, and windows that start at the same moment take turns through a small locked file, so each gets its own.
  • Keeps the colors out of Git. A settings file it creates goes in .git/info/exclude; a committed one gets a local clean filter that hides Nameplate's lines from git status, git diff and every commit.

VS Code Marketplace ↗ Open VSX ↗ GitHub repo ↗ How it works

flakestat

Sep 2026 · Go, zero dependencies · Homebrew, npm, PyPI, Docker, GitHub Action

An open-source flaky test detector for any language. One static Go binary, no SaaS account, and test results never leave your machine.

test runs on commit 4be1f09 score verdict confidence
Fig. 5 A toy version of the idea, not flakestat's real scoring. Score is how inconsistent a test is on one commit, so a test that fails every time scores zero: it isn't flaky, it's broken. Confidence only grows with evidence.

Problem statement: a flaky test passes and fails on the same code. The good tooling for finding them is hosted and paid; the open-source alternatives are a scatter of per-language plugins.

  • Reads the JUnit XML every major test runner already writes, so it works with pytest, Jest, go test, RSpec, PHPUnit and the rest with no plugin and no framework lock-in.
  • Scores inconsistency rather than failure rate. A test that fails every time isn't flaky, it's broken, so it scores zero and is reported separately instead of topping the list and wasting the time of whoever is hunting nondeterminism.
  • Every verdict carries a confidence level derived from how much evidence exists, and classification uses a lower bound on the score, so a small sample can never claim high confidence no matter how flaky a test looks.
  • Validated against three external projects under a contract written before any experiment ran. The strongest result: at the commit before ConduitIO's own deflaking fix, three of the four tests their issue named scored flaky at 0.67 with high confidence; at the fix itself, all four were stable. Their contributors found the bug and wrote the repair — flakestat was handed observations from both sides and separated them.

GitHub repo ↗ flakestat.com ↗

Earlier work

  1. A CNN trained to classify the 7 types of Indian paper currency, served through a Flask app with CSRF protection and deployed on Heroku.

    Date
    June 2020
    Algorithms
    Convolutional Neural Network (CNN)
    Technology
    Python, Jupyter Notebook, Flask

    Problem statement: build an app to predict Indian paper currency.

    • Collected images from search engines and trained a CNN to predict the 7 types of Indian paper currency: 10, 20, 50, 100, 200, 500 and 2000.
    • Deployed the model on Heroku with a Flask back-end and secured the app with CSRF protection.

    GitHub repo ↗ Live demo ↗

  2. Date
    Oct 2017 – Mar 2018
    Client
    Personal
    Technology
    Hadoop, Flume, Hive, Twitter Streaming API, Python

    Twitter, one of the largest social media sites, receives tweets in millions every day. This huge amount of raw data can be used for industrial or business purposes by organizing it according to requirement and processing it. This project provides a way of doing sentiment analysis using Hadoop, which processes the huge amount of data on a Hadoop cluster faster and in real time.

  3. An open-source tool built on the GitHub REST API v3 that tells contributors where they stand on their four Hacktoberfest pull requests.

    Date
    October 2017
    Technology
    HTML, CSS, JavaScript, Bootstrap, Git REST API v3
    Contributors
    See complete list
    • Hacktoberfest is a month-long celebration of open source software in partnership with GitHub, in which participants need to make 4 pull requests across GitHub.
    • Hacktoberfest Status Checker is an open-source tool to know the status of your Hacktoberfest activities during October.

    openstalk.tech ↗

  4. Date
    September 2017
    Technology
    HTML, CSS, jQuery, PHP, Java
    Platform
    Web, PWA, Android
    Contributors
    Shriom Tripathi, Soumyajit Dutta, Biswarup Benerjee
    • Smart Q-Labs is a dynamic queue management solution that takes care of your queue number and notifies you from time to time, plus analytics for outlets so they can manage, and enjoy managing, their queues.
    • I designed the mobile website using the concept of PWA (Progressive Web Apps), specially built to work offline or on a bad network connection. PWA uses modern web capabilities to deliver an app-like experience, using the app-shell model for app-style navigation and interactions.

    GitHub repo ↗

  5. A URL shortening service hosted on AWS with a Flask back-end, built under the Microsoft Technical Community at LPU.

    Technology
    HTML, CSS, JavaScript, Flask, Bootstrap
    Contributors
    Soumyajit Dutta, Biswarup Benerjee
    Made under
    Microsoft Technical Community, LPU
    • The project centres around developing a URL shortening service. You have long URLs that are hard to remember, so use shortTo.com to shorten them into something easy to recall.
    • Hosted on AWS with Flask in the back-end. The Bootstrap framework and JavaScript with media queries make the UX/UI responsive and interactive.

    www.shortto.com ↗

§5

Writing & research

Long reads on the kind of work I do and on the problems inside my own products, each with figures you can push around. Below them, the papers.

Engineering notes

  1. 01A test that always fails isn't flakyTesting · 11 minFailure rate ranks the wrong tests. What flakestat measures instead, why its verdicts wait for evidence, and the times it fooled itself before it could fool anyone else.
  2. 02Ten times faster, mostly by doing lessData engineering · 8 minI rebuilt a graph engine from R to Python and Polars, and it got ten times faster on 80% less memory. Very little of that came from changing languages.
  3. 03Push, don't pollReal-time systems · 9 minGetting a notification from a background job to a browser in under 100 milliseconds is easy once. The hard parts arrive when it has to happen for everyone at once.
  4. 04Spot instances without the 3am pageCloud infrastructure · 8 minSpare cloud capacity is cheap until it's taken back. Interruptions, checkpoints, retries and the spot/on-demand mix behind a batch pipeline bill that came down by 28%.
  5. 05Drawing a graph that's too big to drawVisualisation · 8 minSpatial indexes, label placement, Barnes–Hut, culling, level of detail and a single event listener: the unglamorous techniques that make a huge biological network feel instant.
  6. 06Retrieval is the hard partGenAI · 8 minWhen a RAG system gives a bad answer, the model usually takes the blame. More often it was handed the wrong context. Embeddings, chunking, approximate search, hybrid ranking and reranking, seen from the retrieval side.
  7. 07A dataset is a productKaggle · 8 minFive datasets and six notebooks on Kaggle, downloaded nearly 18,000 times and built on in 63 public notebooks. What it takes to make data that strangers can trust, from every NeurIPS paper to two million Indian companies, and the notebooks I built on top of my own datasets.

Building my own products

  1. 08Counting days the way the government doesCitizenDays · 13 minAustralia, Canada, the UK, the US and the Schengen area all limit how long you can be away, and no two count a day the same way. How CitizenDays turns trips into days, checks rolling windows without counting day by day, plans trips before they're booked, and is a day stricter than one official calculator on purpose.
  2. 09A ledger its own server can't readClearOwe · 11 minClearOwe keeps track of money between people. It works offline, follows you to a new phone and shares a balance by link, while the server in the middle learns almost nothing. Integer money, a parser for how people really talk about debts, sync that doesn't trust clocks, and a key that lives after the # in a URL.
  3. 10Marked like the real thingPart 107 · ASVAB · CitizenOz · 10 minThree exam-prep apps, for the FAA drone licence, the US military's enlistment test and the Australian citizenship test, run on one engine. What it takes for practice to predict the real exam honestly: tests shaped like the real one, marked by its rules, a syllabus with no gaps, and a readiness score that's hard to flatter.
  4. 11A WhatsApp export is not a file formatYAPD · 15 minYAPD turns an exported chat into a Wrapped-style recap. Reading a format nobody designed, defining what a conversation is, letting a language model quote people without misquoting them, and sealing everything that leaves the device so that not even the server can read it.
  5. 12Which project is this window?Nameplate · 9 minEvery VS Code window looks the same, which is how a command ends up in the wrong terminal. Nameplate gives each project a name and a color of its own in the status bar, keeps the colors of open windows apart, and keeps them out of Git.

All writing →

Peer-reviewed research

Four peer-reviewed papers across IEEE, ACM and international journals covering deep learning, NLP and computer vision, plus a 2025 ASMS poster co-authored with Yale University.

  1. [1]

    MAD-Lib: Library Optimisation for Enhanced Metabolite Annotation.

    Poster TP 476, ASMS 2025, co-authored with Yale University. Cut manual LC-MS annotation from weeks to days.

  2. [2]
  3. [3]

    Driver Fatigue Detection System with Mobile Notification Alarm.

    International Journal of Emerging Technologies and Innovative Research (jetir.org), 5(12), 598–605. ISSN 2349-5162. PDF ↗

  4. [4]

    Sentiment Analysis with Fully Supervised Speaker Diarization.

    Think India Journal, 22(3), 8382–8391. ISSN 0971-1260. ResearchGate ↗

  5. [5]

    Automated Web Development: Theme Detection and Code Generation Using Mix-NLP.

    ACM International Conference Proceedings Series (June 2019). DOI 10.1145/3339311.3339356. ResearchGate ↗

Earlier writing, on Medium

§6

Education & recognition

Where I studied

  1. Aug 2018 – Jun 2019 Data Scientist Nanodegree, Udacity. Verify credential ↗
  2. Aug 2016 – Aug 2020 B.Tech, Computer Science & Engineering, Lovely Professional University.
  3. Mar 2008 – Mar 2016 5th – 12th standard, BSJD Convent School.

Certifications & hackathons

  1. Top 0.05% Kaggle Expert Tier. Datasets (top 0.05%) & Notebooks (top 0.07%). View profile ↗
  2. Winner Infosys Hackathon 2.0, Chandigarh DC. View certificate ↗
  3. Top 10 of 45+ HaXplore Hackathon, IIT-BHU.
  4. 250+ solved DSA problems on LeetCode, Codeforces & CodeChef.
  5. Rajasthan IT Hackathon 2.0, Government of Rajasthan. View certificate ↗
  6. Rajasthan IT Hackathon 4.0, Government of Rajasthan. View certificate ↗

§7

Correspondence

Open to conversations about tech lead roles, architecture and interesting engineering problems. The fastest way to reach me is email.

rohitswamicode@gmail.com
Phone
+91 80594 59498
LinkedIn
linkedin.com/in/rowhitswami ↗
GitHub
github.com/rowhitswami ↗
Kaggle
kaggle.com/rowhitswami ↗
Medium
medium.com/@rowhitswami ↗
Resume
resume.rohitswami.com ↗
Location
India.
QR code to chat with Rohit Swami on WhatsApp
Scan to chat on WhatsApp

Keyboard

1–7
Jump to a section
0
Back to the top
t
Toggle 3am mode
`
Open the shell
Esc
Close whatever is open