Skip to content
View Rupali59's full-sized avatar

Block or report Rupali59

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Rupali59/README.md

Rupali Bhatnagar

Senior Software Engineer · India · open to relocation

tathya.dev · LinkedIn · g.dev/rupali59 · Credly · rupalibhatnagar0509@gmail.com

Nine years building high-scale backend and platform systems. Six of them at ServiceNow across Security Operations and Vulnerability Response, where I shipped engines that route 1M+ vulnerable items a day and dashboards used by 50+ enterprise security teams.

Since 2025 I've been building independently: a Go microservices platform, agent-facing developer tooling shipped as a Claude Code plugin, and production web products. My interests are distributed systems, developer tooling, and the unglamorous problem of keeping large codebases honest.


Experience

ServiceNow — Senior Software Engineer

Mar 2022 – May 2025 · Security Operations & Vulnerability Response · Remote

  • Dynamic Approval Rule Engine — multi-level approval workflow for vulnerability deferment and exception requests, built on Flow Designer to stay configurable across differing enterprise policies. 40% faster approvals.
  • Penetration Testing Request Catalog & Workflow Automation — catalog and automated workflows for pen-test requests across 50+ subsystems, cutting manual intervention by 30%.
  • Background Job Processing Framework — multithreaded framework for long-running work that previously strained system resources, with job status tracking, cancellation and ETA visibility. Up to 60% performance gain, 1M+ tasks/month.
  • Real-Time Data Aggregation Engine — high-performance aggregation for near real-time multi-dimensional analytics, processing 100K+ records per 8-hour cycle to power internal dashboards.

ServiceNow — Software Engineer

Apr 2019 – Mar 2022 · Security Operations & Vulnerability Response · Hybrid

  • Assignment Rule Engine — cache-optimised engine prioritising infrastructure vulnerabilities and policy violations and routing them to the right remediation groups. Runs daily over 1M+ vulnerable items.
  • ML-Based Assignment Recommendations — classification-based recommendation system with feature engineering and a user-facing review interface. +20% task allocation accuracy.
  • CISO & Vulnerability Dashboards — led design and development of the executive CISO dashboard and its supporting vulnerability dashboards, defining the key metrics and the data pipeline behind them. Adopted by 50+ organisations.
  • Vulnerability Source Integrations — integrations with Veracode, Qualys, Tenable and Microsoft TVM, standardising 500K+ vulnerability records/day into a common model for downstream remediation.
  • Hackathon projects — Diagnostic Framework (40% less debugging time) · ML vulnerability grouping and remediation-target estimation · PoA&M generator · High-volume data generator (100M+ records across 20–30 tables, adopted by the performance team).
Earlier roles — Darwinbox · NowFloats · Finomena · Babajob (2016–2019)

Darwinbox — Software Developer

Aug 2018 – Mar 2019 · Hyderabad

  • Audit-logging microservice capturing every client data modification — PHP/Yii listening to MongoDB oplogs and publishing to AWS SQS for downstream processing.
  • Oplog diff processor resolving changes across strings, arrays and nested documents into human-readable audit entries.

NowFloats / Boost360 — Software Engineer

Jun 2017 – Aug 2018 · Hyderabad

  • Responsive web platform features supporting 50,000+ daily active users (C#, Bootstrap, JavaScript, jQuery, Knockout.js).
  • SPIDER sales-prediction model — per-representative revenue forecasting at 84% accuracy.
  • Generic booking system on Node.js, configurable across hotels, clinics and restaurants.
  • Facebook Graph API integration for merchant product sharing; order management dashboard (AngularJS, C#).

Finomena — Software Developer

Feb 2017 – May 2017 · Bengaluru

  • Loan management, repayments and collections dashboards in PHP/Laravel, scaled to 8,000+ loans/month on MySQL via Eloquent.

Babajob — Software Engineer

May 2016 – Jan 2017 · Bengaluru

  • IVR and SMS system in Go with RabbitMQ — asynchronous, state-driven inbound call handling at 10K+ calls and 20K+ SMS per day, with MongoDB-backed analytics.
  • Search subsystem in Java over Elasticsearch, with dynamic query generation via Mustache templates.

Projects

Current · 2025 – present

My main line of work is a documentation-integrity toolchain — tooling that detects when files which must agree have silently stopped agreeing — running across every workspace in the tree and enforced by commit and session hooks.

Project What it is Stack
propagate — a Claude Code plugin Agent-facing developer tooling: 4 slash commands, a SessionStart hook that injects canonical engineering rules into every session, and a PreToolUse guardrail that inspects the literal command an agent is about to run and surfaces the matching known hazard before it executes — hazard docs are pull artifacts, and the failure mode is recognition rather than knowledge. Underneath: declares couplings between files that must change together and derives drift from file content on demand rather than storing state. It never edits a downstream file — the tool finds and explains, a human approves, and the decision is recorded in an append-only ledger. Every detector and every guardrail trigger ships with a selftest asserting it can actually fire, because a check that cannot fail reports success — that selftest has caught real bugs. Rearchitected across two generations to an immutable, ULID-keyed, content-pinned store; made portable after the suite passed locally and failed on every clean machine. 130 test files. Node · ESM · Claude Code plugin API
Multi-tenant CRM & workflow platform Express API plus two Next.js 16 apps (client and admin). A DB-driven RBAC/ACL engine — runtime-editable roles, generic field-level access control over CRUD, per-actor tier projection, table→field cascade, and a capability-matrix UI whose review-diff modal shows the consequences of a change before an atomic batch save, with fail-closed reconciliation and admin-floor guards. WebAuthn + OTP auth, BullMQ/Redis job queue, workflow engine, web-push. A phased task-model migration dark-shipped in five stages with a one-shot backfill of ~14,000 records. Test infrastructure cut from 78s to 28s by running one shared mongod for the suite instead of 89. 228 test files. Express · BullMQ · Redis · MongoDB · Next.js · React 19
Content & SEO platform Production site backed by the computation service below — the two are one system. Ten templated landing pages on a shared shell with FAQPage structured data and selective noindex, a server-side analytics pipeline, signed-cookie location handling with a durable resolved-location cache, and a migration of public pages to static generation. Next.js · TypeScript · Radix · CASL
Motherboard Go multi-module monorepo: an API gateway and multi-tenant plugin platform. 36 go.work modules, 702 Go files. Plugin services authenticate, publish their routes to a client manager, and the gateway proxies to them behind a fail-closed entitlement gate and a plugin-key claim assertion. Go 1.25 · Next.js
Demand-intelligence platform Multi-source ingestion and opportunity scoring, with a production LLM classification pipeline: Gemini calls constrained by enforced response schemas, run at temperature: 0 and batched, with determinism resting on cache-on-first-write rather than on the model — and a keyword-classifier fallback so a provider outage degrades quality instead of breaking ingestion. Model-proposed sources are existence-checked against the live API before being accepted, and an inconclusive check returns the candidate flagged rather than asserted either way. Ingestion also covers Search Console via Workload Identity Federation — adopted to remove service-account JSON-key fragility — plus Trends and YouTube. Entity ontology mapping demand against supply; scheduling moved off hosted cron onto a local agent. Next.js · TypeScript · Mongoose · MongoDB · Gemini
Ephemeris computation service FastAPI service over Swiss Ephemeris, a native library with process-global mutable state that is not thread-safe. Every call is serialised through a single-worker executor that releases the event loop during computation — the actual engineering problem behind the API. Adds an astronomical-timeline subsystem (boundary and instant finders, lunar-month tables, per-year caching with TTL) and typed per-calculation endpoints replacing one overloaded route. 1,000+ tests. Python 3.12 · FastAPI · MongoDB
E-commerce storefront & embedded admin app Shopify Liquid theme plus an embedded Remix admin — metafield/metaobject schema installer, headless catalogue seeders, and certificate verification implemented natively in Liquid rather than through an app proxy. Accessibility pass to a 44×44 touch-target floor. Shopify · Liquid · Remix · Polaris · Prisma
Text digitisation & translation pipeline Ingestion for a structured-JSON corpus of classical texts: a TEI/SARIT parser handling prose and division-type detection, a scraper that splits on the source's own headings, a verse-marker grammar, per-work structural maps and validators. One structural fix took depth errors in a single work from 3,522 to 6. Corpus: Sanskrit-texts. Python · JSON
Filmmaker portfolio Production site with a Google Docs → works sync pipeline and multi-codec delivery (AV1/HEVC/H.264): a hero codec ladder took 10.4 MB to ~3 MB, with delivery suppressed for non-watchers, and an H.264 Level 4.0 re-encode for iOS autoplay. Next.js · React 19 · MongoDB
Job-radar skill A Claude Cowork skill that categorises job postings, parses compensation and produces a match-score breakdown, with a local server and a health model. Tested against fixtures and a golden set. Node · esbuild · vitest
tathya.dev My own practice — business sites and CRM. WebAuthn, CASL authorisation, d3-based assessment engine. Next.js 16 · React 19

Also in this period: a Telegram bot service (grammY + webhook dispatch, inline-keyboard state machine, adaptive heartbeat and briefing schedulers, en/hi i18n) and a ~30-page production site rebuilt in Next.js against a live reference.

A 154-commit recovery. A worktree removal combined with dropped stashes left 154 commits unreachable in a live repository. All 154 were located, recovered and permanently tagged, and the restored work was reintegrated behind its tests.

AI & agent tooling

Everything here I built; each is a repository of mine.

  • A production LLM pipeline, engineered for unreliability. Enforced response schemas, temperature: 0, batching, a non-LLM fallback path so an outage degrades quality without stopping ingestion, and containment for the failure that matters most: a hallucinated source is an invisible absence that reads as coverage, so every model-proposed source is existence-checked before acceptance and an inconclusive check is flagged, never guessed.
  • A Claude Code plugin and a plugin marketplace. propagate ships 4 slash commands, 3 hook scripts across 4 registrations on 2 events, 2 bundled skills, and a skill lifecycle CLI (create / promote / demote / reap); the marketplace serves 19 skills across 3 plugins.
  • An agent guardrail. A PreToolUse hook that pattern-matches the command an agent is about to execute against declared hazards and delivers the relevant one at the moment of risk — written after a documented hazard, sitting three files away, was hit anyway and cost 11 false verification records.
  • Evaluation harnesses, not vibes. Detectors and guardrail triggers each carry a selftest asserting the check can fire against a known-positive input; the job-radar skill is tested against fixtures and a golden set. A check that cannot fail is worse than no check, because it reports success.
  • Human-in-the-loop by construction. The tooling is designed so an agent proposes and a person disposes: no downstream file is ever edited automatically, and every approval is an append-only ledger event.

Client work is described by what was built rather than by client name. Those repositories are private; the products are in production.

Earlier

Project
HIMK — human activity recognition using an HMM-based Intermediate Matching Kernel with SVM. M.Tech thesis, re-implemented in 2026 as a containerised Python pipeline with a Flask front end Python · scikit-learn · hmmlearn · OpenCV · Docker
MIS — institute management platform for students, faculty and administration Next.js · MongoDB
Parking Lot — system-design allocation problem Go
WorkTracker — personal work dashboard and digital garden TypeScript
Voice Transcriber — dockerised real-time speech-to-text (private) Python · Docker

Technical Skills

Languages Go · TypeScript · JavaScript · Python · Java · PHP · C# · C++
Web Next.js · React · Node.js · Remix · Astro · Laravel · .NET · Tailwind
Data PostgreSQL · MongoDB · MySQL · Redis · Elasticsearch
Messaging RabbitMQ · BullMQ · AWS SQS
Commerce Shopify · Liquid · Polaris · Prisma
Cloud & tooling AWS · Vercel · Docker · Fly.io · Git · Linux
AI & agent tooling Claude Code plugins, hooks and skills · structured LLM output and schema-constrained generation · eval harnesses and golden sets · hallucination containment and graceful degradation · human-in-the-loop design
ML scikit-learn · hmmlearn · OpenCV — pattern recognition and classification (thesis; ServiceNow assignment recommender)
Platform ServiceNow SecOps · Vulnerability Response · GlidePlatform · Flow Designer

Education

Degree Institution Grade Years
M.Tech, Computer Science & Engineering — Pattern Recognition & Machine Learning National Institute of Technology Goa 8.42 2014–2016
B.E., Computer Science & Engineering LNCT Bhopal (RGPV) 8.58 CGPA 2010–2014

Certifications

AI, 2026 — Anthropic: 5 certifications across Claude, Claude Code and Claude Cowork (Jul–Aug 2026). Google: Gemini Enterprise Agent Ready · Create Your First Gemini Enterprise Application · Google Cloud Innovator · Developer Program Premium Tier · AI Fundamentals (Google Career Certificates).

Earlier — WhatsApp API Expert, Wati (2024) · Ultimate JavaScript: Fundamentals, Code With Mosh (2025) · Neural Networks & Deep Learning, Coursera (2020) · JavaScript Algorithms & Data Structures, freeCodeCamp (2020) · Git Essential Training, LinkedIn Learning (2020)


Honors & Awards

  • Winner — NowFloats internal hackathon, for the SPIDER sales-prediction module
  • Chancellor's Scholarship — top 10 students statewide by theory marks, RGPV Bhopal · 2012 and 2013
  • Silver Medal — Persistent Systems Programming Contest, inter-college · 2013
  • Teaching Assistant, Computer Science Department — NIT Goa · 2014–2016
  • Vice President, Computer Science Department — NIT Goa · 2014–2016

Open to Opportunities

Senior and staff-level platform, backend and full-stack roles — remote or on-site. Also open to open-source collaboration on developer tooling. Email me or reach out on LinkedIn.

Pinned Loading

  1. Leetcode Leetcode Public

    Leetcode solutions

    JavaScript 1

  2. geektrust-problems geektrust-problems Public

    GeekTrst Problem Solutions

    JavaScript 1