Systems · Databases · Observability

Vanshika
Turakhia

I build systems that move data quickly, stay correct under pressure, and explain what happened when they don't.
CS student focused on C++, databases, concurrency, and backend infrastructure.
scroll
Hardware, though?

Most of what's above lives on servers. This part you can touch.

Point-to-listen museum case

Museum exhibit · in progress

A display case you talk to by pointing. Hold a finger over an object for three seconds and its story plays in the headphones, while the screen under the glass glows beneath it. No buttons, no touchscreen: a camera above the case watches for your finger, and a passing hand does nothing.

Raspberry Pi 4 · Camera Module 3 NoIR · Python · OpenCV · background subtraction, fingertip = the point reaching farthest in · camera-to-TV calibration · ~11 fps at 960×540
Research

Questions I'm chasing that don't have answers yet.

Keeping urgent jobs on time

Real-time scheduling · in progress

When a small computer runs a machine, some jobs can't be late: a motor controller has to answer on time, every time. I'm testing whether a different way of letting urgent jobs cut the line makes that timing steadier on a Raspberry Pi, compared with the tricks people already use.

FreeRTOS · AArch64 on Raspberry Pi 4 · preemption threshold vs temporary priority raise vs scheduler suspension · cycle-counter instrumentation for latency and missed deadlines · hardware runs pending

Editing a tree that won't sit still

Concurrent data structures · in progress

Picture a huge family tree or folder structure that lots of people change at once. DomLock is an existing way to let them work without stepping on each other. I'm studying how to apply many changes together in one batch, safely, while the shape of the tree itself keeps changing.

C++20 · DomLock · batched edge insert/delete on a DAG · prepare / validate / merge / apply deltas · CSR engine run at 1M–10M nodes · exact model and Python oracle as cross-checks

When the shortcut gets slow

Data structures · in progress

Lots of software keeps a running list and constantly asks "what's next?", like a server tracking which connections have timed out. The standard tool for this uses a clever shortcut that's usually faster. I found a pattern where it gets over five times slower, and it's the pattern those servers actually create. I'm working out exactly why.

C++20 · d-ary heap from scratch vs libc++ std::priority_queue · bottom-up (Floyd) pop 5.6× slower at 16M when pushes land near the top · settle depth ≈ log₂S / log₂N · algorithm and API effects measured separately
The short version

I like building interactive things, but not everyone has time to explore a portfolio. Here is the short version:

  • I build C++ systems projects around event loops, concurrency, memory, and database execution.
  • I have built a vectorized query engine that showed a 12.7x improvement over an iterator baseline on the same workload.
  • I have built an observability prototype using Kafka, ClickHouse, FastAPI, and Grafana.
  • I contribute to open source systems software including Valkey and KubeVela.
  • I take on small, clearly scoped backend, data, integration, and reliability engagements.
Open Source

Valkey

Contributor · C

Implemented a trusted connection-pool path so designated administrative and monitoring clients can still connect when maxclients is exhausted, avoiding an operational deadlock during incidents.

KubeVela

Contributor · Go · Kubernetes

Added lifecycle-management APIs for explicit goroutine shutdown in long-running controller processes. Also shortlisted for a CNCF LFX mentorship project around LRU eviction for native Helm providers.

Competitive Programming
1604
CF max rating · Specialist
300+
Problems solved
About

I'm Vanshika, a CS student who likes the part of software where tradeoffs are real: concurrency, databases, systems performance, observability, and infrastructure.

I study Computer Science at BITS Pilani (B.Sc., expected 2029) and am based in Bengaluru, India.

I learn best by building close to the metal, measuring behaviour, finding the failure mode, and then making the design simpler. I'm currently looking to grow through systems-oriented internships, open source, and carefully scoped engineering work.

Stack
LanguagesC++20 · Go · Python · TypeScript · SQL
Systemsepoll · mmap · sendfile · std::atomic · ASan · TSan · perf
InfraKafka · ClickHouse · PostgreSQL · Redis · Docker · Grafana
AlsoAPIs · integrations · data ingestion · observability
View source