Open to Software Engineer roles

Oitash Ghosh

I build systems that stay correct when things break. Consensus, storage engines, and tests that try to break them.

Salt Lake, Kolkata, India oitashg@gmail.com
Portrait of Oitash Ghosh
Currently
HandshakeAI Evaluation Specialist
Previously
AmazonData Engineer Intern
Education
Jadavpur UniversityB.E. Information Technology
01 / About

Three disciplines, one engineer.

I graduated in Information Technology from Jadavpur University, and I'm drawn to the parts of software where correctness is hard to fake: distributed consensus, storage engines, and hot paths where nanoseconds show up in a benchmark.

At Amazon I re-architected an inventory audit pipeline to 100% parity with the legacy stack and shipped a Python CLI that cut deployment time by 40%. On my own time I've written a Raft-replicated key-value store with its own LSM engine and linearizability checker, and a matching engine that clears a message in 25 nanoseconds.

I'm currently open to Software Engineer roles.

Distributed systems

Consensus and storage built from first principles: a Raft-replicated key-value store with its own LSM engine, checked for linearizability under injected network partitions.

Data engineering

Spark and EMR pipelines built for correctness first, with anomaly detection, parity testing against legacy systems, and quality gates that catch bad data before it reaches a dashboard.

Full-stack product

React and Django/Node apps shipped end to end: auth, payments, media, real-time dashboards, and the unglamorous deployment work that makes them stay up.

Fundamentals

1000+ solved problems, a CS core of DSA, DBMS, OS and networks, and a habit of reaching for the simplest structure that holds under load.

02 / Experience

Where I've worked.

From petabyte-scale supply chain pipelines at Amazon to evaluating frontier models on real engineering tasks.

  1. Handshake

    AI Evaluation Specialist

    Jul 2026 — Present · Remote

    Building containerized benchmarks that measure what AI coding agents can actually do.

    • Engineered 40+ containerized evaluation environments with automated graders, benchmarking AI coding agents across concurrency debugging, security, ML infrastructure and low-level systems.
    • Cut agent solve rate from 80% to 40% on a distributed systems task by redesigning it to test derived reasoning rather than pattern recall.
    • Validated each task against 3 independent implementations and proved every grading rule affects 2+ cases, catching a grader defect that was failing correct submissions.
    PythonDockerAutomated GradingAgent EvaluationDistributed Systems
  2. Amazon

    Data Engineer Intern

    Jan 2026 — Jun 2026 · Bengaluru, India

    SCOT-AtmtdInvntryMgmt(fka2151) (5528)

    Built and migrated petabyte-scale inventory pipelines for the global supply chain org.

    Project
    Data Quality and Observability Framework for Large-Scale Data Pipelines Using Apache Spark

    • Designed a Spark/EMR Data Quality Framework processing 1 billion+ daily records across 5 regions, deploying a Z-score anomaly detection model that improved data reliability by 25%.
    • Migrated 10+ big data pipelines, resolving Scala build conflicts, and built a Python CLI to bulk-migrate 100+ profiles, cutting deployment time by 40%.
    • Re-architected a Low Inventory Audit pipeline serving 11 marketplaces and 20+ data sources, achieving 100% output parity with the legacy infrastructure.
    • Developed a 14-day advance notification system for inventory levels, preventing over 25% of critical monthly stockout events globally.
    SparkScalaPythonAWS EMRSQLAirflow
    Internship certificate
03 / Projects

Things I've built.

Full products, not demos. Each one shipped with the constraints, tests and trade-offs that come with real users.

Stratakv

Aug 2026

A linearizable distributed key-value store built on Raft consensus, an LSM storage engine, and a linearizability checker that verifies the whole thing under network partitions.

  • From-scratch Raft covering leader election, log replication, current-term-only commitment and snapshot compaction, sustaining 1,100 ops/s across 3 nodes, with leader failover measured at 334–526 ms.
  • Hand-written LSM storage engine: a CRC-framed write-ahead log with torn-tail recovery, skiplist memtable, bloom-filtered SSTables and levelled compaction. Point reads land at 1.17 µs, and the filters skip 99% of tables on absent keys.
  • Correctness is proved rather than asserted. A self-written linearizability checker (Wing & Gong, with Lowe's memoisation) verified 3,150 operations across 15 leader changes under repeated partitions, with zero anomalies. The checker itself is tested against six classes of known violation.
  • Raised throughput 25 → 1,100 ops/s by isolating each layer: group-commit durability, an append-only log, and a lock-free counter on the apply path that was serialising every operation behind a 4 ms fsync.
GoRaftLSM-TreeDistributed SystemsDocker

At a glance

3,150
ops verified linearizable
334ms
fastest failover
0
external dependencies

Order Book Matching Engine

Aug 2026

A price-time-priority limit order book built the way a trading venue's hot path actually is: flat arrays instead of node graphs, a pre-allocated pool instead of malloc, a bitmap instead of a tree walk.

  • Sustains 25 ns per message (≈39M msg/s, single-threaded) at p99 125 ns, using flat price-level arrays, intrusive FIFOs and a lock-free feed handoff.
  • Validated trade-for-trade against an independently written reference implementation over 12M+ randomised commands, with zero divergences.
  • Clean under AddressSanitizer, UndefinedBehaviorSanitizer and ThreadSanitizer. The benchmark probes the clock's own cost and resolution first, so a 25 ns operation is never measured through a 14 ns observer effect.
C++20MultithreadingCMakePython

At a glance

25 ns
per message
125 ns
p99 latency
12M+
commands, 0 divergences

ELD Trip Planner

Jul 2026

A full-stack hours-of-service planner that turns four trip inputs into a compliant route and auto-generated FMCSA log sheets.

  • Renders a 24-hour duty grid per day in SVG, auto-generating the FMCSA log sheets drivers are legally required to keep.
  • Dependency-free rules engine encoding six federal driving and rest limits, backed by 20 unit tests and a self-validation pass that flags violations.
  • Replaced rate-limited reverse-geocoding calls with an offline 30,000-city index; Nominatim, OSRM and Leaflet give a zero-API-key map stack.
ReactDjangoDjango REST FrameworkPythonPostgreSQLLeaflet

At a glance

6
federal rules encoded
30k
offline city index
0
API keys required

CodeHolic

Dec 2024

A role-based EdTech platform on the MERN stack, with course authoring, progress tracking, quizzes and payments.

  • Three distinct dashboards for student, instructor and admin, with Redux holding global state across role-specific views and permissions.
  • 50 REST endpoints over 12 data models covering courses, sections, quizzes, progress tracking and reviews.
  • Razorpay payments and Cloudinary media behind JWT authentication, with tiered rate limiting on auth, email and global traffic.
MongoDBExpressReactReduxNode.jsCloudinaryRazorpay

At a glance

3
role-based dashboards
50
REST endpoints
12
data models
04 / Skills

The toolkit.

What I reach for day to day, weighted toward systems and backend, with enough front-end to ship the whole thing.

Languages

  • Python
  • Go
  • C++
  • SQL
  • Java
  • JavaScript
  • TypeScript
  • Scala
  • C

Data & Cloud

  • Apache Spark
  • AWS EMR
  • AWS S3
  • PostgreSQL
  • MongoDB
  • Data Modelling
  • ETL / ELT
  • Gunicorn

Web & Frameworks

  • React
  • Next.js
  • Django
  • Django REST Framework
  • Redux
  • Node.js
  • Express
  • Tailwind CSS
  • shadcn/ui

Systems & Distributed

  • Raft / Consensus
  • LSM Storage Engines
  • Concurrency
  • Linearizability Testing
  • Docker
  • CI/CD (GitHub Actions)
  • Linux

Foundations

  • Data Structures & Algorithms
  • DBMS
  • Operating Systems
  • Computer Networks
  • OOP
  • Git & GitHub
05 / Beyond

Education & achievements.

Jadavpur University

2022 — 2026 · Kolkata, India

B.E. in Information Technology

CGPA 7.89 / 10

Relevant coursework

  • Data Structures & Algorithms
  • Database Management Systems
  • Object Oriented Programming
  • Operating Systems
  • Computer Networks

Adobe India Hackathon 2025, Round 2

Hackathon

Qualified for Round 2 building an AI-powered document intelligence solution under strict offline and performance constraints.

1000+ problems solved

Competitive Programming

Across LeetCode, GeeksforGeeks, CodeChef and Codeforces, with sustained practice in algorithms and data structures.

AIR 8120 in JEE Mains

Entrance Exam

Scored 99.1 percentile in the JEE Mains examination.

Rank 566 in WBJEE

Entrance Exam

Top 0.6% out of 100,000+ candidates.

06 / Contact

Let's build something that scales.

I'm actively looking for Software Engineer roles. If you think there's a fit, or you just want to talk about Spark, rules engines or hard bugs, my inbox is open.

+91 91635 85861