Backend · Cloud · Data  /  CSE '27, Sahyadri College of Engineering, Mangaluru

I build systems that have to work.

Final-year computer science student working across cloud infrastructure, data pipelines and backend systems — a carbon-aware cloud scheduler, production web platforms, a data lakehouse pipeline, developer tools people actually use.

See my work Download resume Get in touch

Hover to reconstruct

About me

Systems before slogans.

I'm Nishit — final-year CSE at Sahyadri College, building backends and cloud systems that have to hold up under real constraints, not just demo well. Cloud, data, backend — that's the stack I actually want to work in.

Every project on this page is one I finished and can defend line by line, not another item on a longer list of things I started.

Experience

Internships

Student Intern

Urudha

May 2026 — Jul 2026

Web design & software development agency, Mangaluru

  • Shipped three production websites with Next.js and React, all live today.
  • huskybusky.in — brand platform for eco-friendly tableware made from rice and coffee husk; built frontend features and reusable UI components for product, B2B and brochure pages.
  • houseofjn.store — e-commerce storefront for a Mangaluru handbag boutique: filterable product catalog, collection highlights and WhatsApp-based ordering.
  • squish.urudha.com — the in-browser WebP converter and image toolkit featured below.
  • Collaborated through Git-based workflows to ship responsive, maintainable interfaces.
Next.jsReactGit

Selected Work

Shipped & running

Product · WebP converter

Squish

A WebP converter that runs entirely in the browser — images never leave the device. Converts JPG, PNG, HEIC, AVIF and more into WebP files 25–34% smaller than JPEG, handling single files, folders or whole ZIP archives. Grew into a seven-tool suite: images-to-PDF, video frame extraction, AI background removal, EXIF stripping, palette extraction and batch watermarking — all client-side, live on my own domain.

Client-side processing · Zero uploads · squish.urudha.com

Live ↗ Repo ↗

Data Engineering · ETL

Airflow Medallion Pipeline

A medallion-architecture ETL pipeline: raw data lands in Bronze, gets cleaned and conformed in Silver, and is aggregated into analytics-ready Gold tables. Transformations run on Apache Spark, orchestrated end-to-end by Airflow DAGs with Delta Lake for versioned storage — the same layered pattern used in production lakehouses.

Spark · PySpark · Airflow · Delta Lake

Repo ↗ Runs on AWS EC2 (spun up on demand)

Full-stack · Computer Vision

Pantry Guardian

Inventory management that reads your groceries for you: OCR ingestion of labels and receipts, shelf-life prediction with expiry alerts, a pantry-health score and recipe recommendations from what's actually on the shelf. Next.js frontend, Express backend, MongoDB storage, Clerk for auth.

Next.js · Express · MongoDB · OCR · Clerk

Live ↗ Repo ↗

Product · Developer tools

Port-Forge

A portfolio generator for developers: sync your GitHub and it assembles a professional portfolio from your real repositories — AI-assisted content, professional templates and real-time insights, so a developer's work speaks before their resume does. Live and open source.

GitHub sync · AI generation · Templates

Live ↗ Repo ↗
Also shipped LearnPath-AI ↗ Everything else on GitHub ↗

Flagship

VTU Major Project · Research Paper In Progress

Scheduler · Forecasting · Cloud

Carbon-Aware AI Workload Scheduler

A two-phase, SLA-first scheduler that decides where and when cloud jobs run. Phase 1 hard-filters regions on latency and capacity; Phase 2 minimises emissions across the survivors using a full energy model (CPU × TDP × exec time × PUE × carbon intensity).

Delay-tolerant jobs get joint spatial + temporal shifting with a 4-hour look-ahead; latency-sensitive jobs shift across space only. Live carbon intensity comes from the Electricity Maps API with MongoDB caching; forecasting benchmarked with LSTM vs ARIMA vs persistence on 4 years of per-zone grid data, tested fully out-of-sample on every season of 2025. An RL scheduler is benchmarked against the rule-based one, with a FastAPI analytics layer and React dashboard on top.

PythonFastAPILSTMARIMARLElectricity Maps APIMongoDBReact
−31.5%CO₂ across all 50 workloads
100%SLA compliance — 50/50 jobs
−39.9%CO₂ on delay-tolerant jobs
6Regions, Norway → Mumbai
Region board — grid carbon intensity, gCO₂/kWh Scheduler input
EU-NORTH-2 · NorwayHydro 97%
~15
EU-NORTH-1 · SwedenHydro + Nuclear
~40
EU-WEST-1 · IrelandWind + Gas
~230
US-EAST-1 · VirginiaMixed
~380
AP-SOUTHEAST-1 · SingaporeGas
~450
AP-SOUTH-1 · MumbaiCoal dominant
~710
Same job, 47× difference in emissions depending on where it runs. That gap is what this scheduler exploits.

Stack

Tools I reach for

Scroll or drag

Languages

  • Python primary
  • Java DSA
  • JavaScript · TypeScript
  • C · C++
  • SQL

Backend

  • FastAPI
  • Node.js · Express
  • MongoDB · MySQL

Data

  • Apache Spark · PySpark
  • Apache Airflow
  • Delta Lake
  • Pandas
  • LSTM · ARIMA

Cloud / DevOps

  • AWS EC2
  • Docker
  • Amazon S3
  • Vercel
  • Git · CI

About

The short version

I'm a final-year CSE student at Sahyadri College of Engineering & Management, Mangaluru, headed for backend, cloud and data roles. I like systems with hard constraints — schedulers that can't miss SLAs, auth that can't leak, pipelines that have to land on time.

I prototype a lot, but the projects on this page are the ones that survived the cut: built, measured and finished. That filter is deliberate — I care more about a system I can defend number-by-number than a long list of half-built demos.

  • DegreeB.E. Computer Science, 2027
  • CGPA9.04 / 10
  • FocusBackend · Cloud · Data Engineering
  • ResearchCarbon-aware scheduling (paper in progress)
  • CertificationsPython & Java Foundation — Infosys Springboard
  • AchievementTop 15 — MAKEFOR Mangaluru (Titan), 2025
  • BaseMangaluru, India

Contact

Open to internships & placements

Have a backend that needs building? Let's talk.

nishitsk42@gmail.com +91 74835 43085 GitHub LinkedIn Resume