Building real-time products and learning how systems behave under load.
View portfolio ↗  ·  LinkedIn  ·  Repositories

I am Gaurav, a software engineering student focused on backend engineering, distributed systems, and practical full-stack development. I build to understand the details: shared state in a collaborative editor, changing ride state across two mobile apps, inventory under contention, and the tools around an AI coding agent.

I completed a three-month Software Engineering Trainee internship at FamilyShell. Contributing to established open-source projects has also taught me to read unfamiliar code, test changes, and work through upstream review.

3 merged upstream  ·  14 open upstream  ·  7 organizations  ·  1 highlighted bug report
GitHub PR states checked 16 October 2026. “Open” means proposed, not merged.

Merged upstream

• apache/beam #40496 — added opt-in lazy Dask side inputs with worker-safe partition computation.
• kubernetes/perf-tests #4182 — clarified pod startup status messages during rolling updates.
• kubernetes-sigs/dra-example-driver #259 — restored CDI specs when prepared DRA claims are recovered from a checkpoint.

Open · Kubernetes ecosystem

• kubernetes/kubernetes #141568 — proposes durable, rate-limited retries for failed taint-eviction pod deletions.
• kubernetes/kubernetes #143105 — proposes the default dialer when connection-rotation receives a nil dial callback.
• kubernetes/perf-tests #4358 — proposes more precise timestamps for ResourceClaim latency measurements.
• kubernetes/perf-tests #4183 — proposes Go runtime memory, GC, and allocation metrics in benchmark results.
• kubernetes/test-infra #38012 — proposes routing triage bug reports to the repository selected by the Repo filter.
• kubernetes/website #57960 — proposes clearer reservedSystemCPUs precedence and incompatible kubelet cgroup settings.

Open · Google Research and GoogleTest

• google-research/google-research #3447 — proposes avoiding gradient mutation in Shampoo SGD grafting.
• google-research/google-research #3449 — proposes correct Shampoo optimizer state initialization on CPU.
• google/googletest #5123 — proposes aggregate skipped-test counts in GoogleTest XML output.
• google/googletest #5125 — proposes correct JSON output for quoted RecordProperty keys.

Open · Apache Beam

• apache/beam #40625 — proposes resolving string annotations in callable type hints.
• apache/beam #40631 — proposes preserving HTTP errors when quota responses have non-JSON bodies.

Open · Redis and Supabase

• redis/redis #15923 — proposes a bounded redis-cli cluster-join wait when cluster bus ports are unreachable.
• supabase/supabase #51838 — proposes preserving existing nonnumeric SQL LIMIT expressions in Supabase Studio.

Highlighted bug report

• Supabase issue #51827 — Table Editor numeric filters round high-precision decimal values: JavaScript Number conversion changes PostgreSQL numeric/decimal filter inputs (open).

Four projects, each centered on a different engineering problem and a concrete system to explore it.

CodeCollab gives developers a shared, multi-file coding workspace. Monaco and Yjs handle collaborative editing; chat, presence, and code execution sit alongside a Next.js frontend, Spring Boot API, and Node.js sync service, with MongoDB and Redis.

Flux coordinates passenger requests, rider bids, and changing ride status across two React Native apps. Its Spring Boot backend uses WebSockets, PostgreSQL, and Redis; the apps contain map and location flows. Flux is a demo/test project, and its current payment service uses free subscriptions.

FlashFlow explores inventory reservation when many requests contend for limited stock. An API gateway separates authentication, stock, and order services; Redis coordinates stock operations, MongoDB stores service data, and Kubernetes manifests plus a k6 script support repeatable load experiments.

NeuraSWE is an AI coding-agent prototype that turns GitHub issues or Jira webhooks into queued runs. A FastAPI service, arq/Redis worker, and ReAct-style runtime connect repository search, file edits, Docker sandbox commands (with a local fallback), test execution, and optional pull-request creation. A React dashboard streams run activity; PostgreSQL stores run state.

Languages   Java · TypeScript · JavaScript · Python · SQL
Backend & real time   Spring Boot · Node.js · Express · FastAPI · WebSockets · Yjs
Data & messaging   PostgreSQL · MongoDB · Redis · arq
Product surfaces   React · Next.js · React Native
Infrastructure & tooling   Docker · Kubernetes · GitHub Actions · k6
Agent tooling explored in NeuraSWE   ChromaDB · Ollama · Docker sandboxing

These third-party cards are optional visuals; the repositories remain the source of truth if a card is unavailable.

View the contribution history directly if the animation is unavailable.

I practice data structures, algorithms, backend fundamentals, and system design. My September 2026 self-reported progress was 343 LeetCode problems and 404 Take U Forward problems. These sets may overlap; they are not a unique total.

Portfolio  ·  LinkedIn  ·  GitHub

For a project-specific discussion, open an issue in the relevant repository.