DZ — Portfolio 2026--:-- IST

Devansh
Zode

Portrait of Devansh Zode

AI full stack developer and final-year AI & ML student. I build backend-heavy systems — real-time multiplayer, event-driven pipelines, computer vision — and the interfaces around them.

Open to software engineering internships

01Work

Synchro

Built

Real-time multiplayer co-op puzzle game

A three-player co-op game on a server-authoritative Socket.IO backend. Each player sees a different screen and has 60 seconds to find the mismatched line — so the server decides roles, views, time and the win, and never trusts the client.

client Aplayer 1inputserverSocket.IOviewclient Bplayer 2

What was actually hard

  • 01

    Server authority

    Role assignment, per-player views and win validation all live on the server, so no client can cheat or drift.

  • 02

    A timer that holds

    A drift-free countdown shared by three players, with automatic reset on wrong lock-ins.

  • 03

    Voice, peer to peer

    WebRTC voice chat in a 3-way mesh, with a custom signaling relay and live teammate selection highlights.

OAuth login, user stats and match history in a TypeScript pnpm monorepo with shared client/server types.

Next.js / TypeScript / Node.js / Express / Socket.IO / WebRTC / PostgreSQL / Prisma / Auth.js
deployed on Vercel + Railway

Source ↗

More projects

Built

Stratify AI

Market intelligence platform

  • Event-driven backend (FastAPI, transactional outbox, RabbitMQ workers) that splits research questions into parallel tasks and writes cited reports with Gemini and Tavily.
  • Multi-agent debate engine grounded in SEC EDGAR, arXiv and GDELT, with recommendations backtested via a Decision Quality Score.
  • Next.js dashboard, 45 REST endpoints, HMAC-signed webhooks, SHA-256-sealed exports and 119 automated tests.

Next.js / TypeScript / Python / FastAPI / PostgreSQL / RabbitMQ / Docker

Source ↗
Building

Inspect AI

Industrial visual anomaly detection

  • Detects and localizes defects in industrial images using anomaly detection, trained on mostly defect-free data.
  • Reconstruction-based (CNN autoencoders) and feature-based (PatchCore) models, compared head to head.
  • Evaluated with image-level and pixel-level metrics, anomaly heatmaps and comparative experiments.

Python / PyTorch / OpenCV / CNNs / Autoencoders / ResNet / PatchCore / MLflow

Source ↗

02Stack

Languages
Java, JavaScript, TypeScript, Python, SQL
Frontend
React, Next.js, HTML5, CSS3, Tailwind CSS
Backend
Node.js, Express.js, FastAPI, REST APIs, Socket.IO, WebRTC, RabbitMQ, Auth.js
AI / ML & vision
PyTorch, OpenCV, CNNs, Autoencoders, ResNet, PatchCore, MLflow
Data & tools
PostgreSQL, Prisma, Docker, Git, GitHub, Vercel, Railway, pnpm
Core
OOP, data structures & algorithms, DBMS, operating systems

03Education

’27

B.E. Artificial Intelligence & Machine Learning

Zeal College of Engineering and Research, Pune

Expected June 2027 · CGPA 8.05 / 10

Coursework

Data Structures & Algorithms · Database Management Systems · Operating Systems · Software Engineering · Computer Networks

04Contact

Hiring interns, or building something real-time?

devanshzode.dev@gmail.com