Projects

A Toolforge tool that finds dates and numbers that disagree between language editions of the same Wikipedia article and turns them into a worklist for editors. It pairs statements across languages through Wikidata, asks an open-weight LLM on LiftWing for conflicting values, then verifies each one and filters out rounding, unit and calendar differences.

  • Python
  • Flask
  • Wikidata
  • LLMs
  • Toolforge

WikiNPOV Drift Detector

In progressT428896

An ML tool that detects neutrality drift in Wikipedia articles across revisions. It tracks nine signals per revision, from subjectivity and attribution patterns to editor concentration and revert rates, and uses changepoint detection (PELT) to flag when drift began, surfacing evidence for human editors rather than making judgments itself.

  • DeBERTa-v3
  • MiniLM
  • Changepoint detection

Wikicap

A media-accessibility platform for Wikimedia Commons with automated transcription, a live visual editor for SRT/VTT subtitle timing and server-side MP4 rendering with Remotion, authenticated through MediaWiki OAuth.

  • React Router v7
  • Node.js
  • PostgreSQL
  • Remotion

Research

Jul 2026 – Present

CSTAR Lab, IIIT Hyderabad

Research Intern · Hyderabad

Researching multi-GPU inference for high-resolution image generation with diffusion models, studying displaced patch parallelism (DistriFusion) and patch-level pipeline parallelism (PipeFusion). Building and profiling PyTorch inference pipelines for latency, throughput and scaling efficiency across GPUs on a single node.

  • PyTorch
  • Diffusion models
  • Distributed inference

Jun 2025 – Aug 2025

Hand Gesture Occlusion

Research Project · Delhi

Built a scalable pipeline that extracts 3D hand landmarks from RGB images and compares gestures with point-cloud similarity metrics, using MediaPipe, Open3D and PyTorch3D (Chamfer distance), with parallel GPU-accelerated processing over large gesture datasets.

  • Computer vision
  • 3D point clouds
  • PyTorch3D

Feb 2024 – Apr 2024

CyPSi Laboratory

Research Intern · Delhi

Developed a retinal image segmentation system for medical diagnostics using U-Net and CNNs on fundus images, with GAN-based data augmentation to improve generalisation.

  • Medical imaging
  • U-Net
  • GANs

Experience

Apr 2025 – Jun 2025

Sarvam AI

Frontend Developer Intern · Bengaluru

Built and maintained real-time AI chat interfaces in React with streaming text, and worked with engineering and product teams on UI/UX, responsiveness and frontend performance across production AI applications.

  • React
  • Streaming UI
  • Performance

Dec 2024 – Feb 2025

LiveBuy

Software Developer · New Delhi

Built the company's web presence from scratch, including a Next.js blogging platform, with responsive and accessible interfaces supporting its rental-accommodation product.

  • Next.js
  • TypeScript
  • PostgreSQL

Recognition

  • Smart India Hackathon, Finalist · 2024 & 2025

    Built platforms for anti-doping awareness and a micro-credential aggregator for the Ministry of Youth Affairs & Sports and the Ministry of Skill Development.

  • Samsung Solve for Tomorrow · 2025

    Top 100 teams overall; top 25 in the Social Change through Space and Tech theme.

  • Tata Technologies InnoVent

    Top 31 of 2,822 teams for a digital twin for battery ageing and reuse optimisation.

Skills & interests

Languages
Python, C++, C, Java, TypeScript, JavaScript, R, MATLAB
ML & data
PyTorch, TensorFlow, Hugging Face, Scikit-Learn, OpenCV, FAISS, NumPy, Pandas
Web
React, Next.js, Node.js, Express, Flask, PostgreSQL, MongoDB
Infrastructure
Docker, AWS S3, Firebase, Git, Toolforge, PAWS, LiftWing
Outside work
Chess (mostly bullet, some blitz and rapid) · Speedcubing (3×3, 2×2, pyraminx, mirror cube and other puzzles) · Japanese (learning one character and phrase at a time)

Contact

Always happy to talk about research, open source or a problem worth building for. Email is the best way to reach me: mishradev222004@gmail.com