AI/ML engineer, web and DevOps engineer

Tushar
Singh

Every build is a small eureka.

Download resumeSee projects

About

B.Tech student in AI and ML at GGSIPU EDC (2025 to 2029). Builds models, the services around them, and the pipelines that ship them, from 3D medical imaging to LLM agents that play games.

Tech stack

Languages
Python, C, C++, Java, Bash, HTML, CSS, JavaScript, TypeScript
ML and AI
Deep learning, computer vision, RAG, agentic AI, LangChain, LangGraph
ML libraries
PyTorch, TensorFlow, Scikit-learn, OpenCV, MediaPipe, NumPy, Pandas, Matplotlib
Web and data
FastAPI, React, Node.js, Express.js, REST APIs, PostgreSQL, MySQL, MongoDB
Cloud and DevOps
Git, GitHub Actions, Linux, AWS, Docker, Kubernetes
Jul to Aug 2026

3D Brain Tumor Segmentation

End-to-end 3D segmentation on BraTS 2019 multi-modal MRI. A 3D U-Net is benchmarked against a Swin-UNETR transformer, then both are served through FastAPI with a Streamlit UI for MRI upload, slice overlays and downloadable masks.

  • 0.85 whole-tumor Dice for the 3D U-Net, against 0.74 for Swin-UNETR
  • Per-class Dice of 0.77, 0.84 and 0.80 on necrotic core, edema and enhancing tumor
  • Class-weighted Dice plus cross-entropy loss for the tumor/background imbalance

Python, TensorFlow, 3D U-Net, Swin-UNETR, FastAPI

View source on GitHub
Nov 2026 to present

Neural Flow Dynamics

A research platform for weather forecasting and disaster decision support. Six pipeline stages run from data acquisition to approval, on versioned interfaces with full data provenance. A ClimODE-inspired neural-ODE forecaster is in design.

  • Idempotent, provenance-tracked ingestion with Alembic migrations
  • Source-feasibility checker for ERA5 and NOAA GFS access and licensing
  • FastAPI, React, PostgreSQL and Redis in Docker Compose, checked by pytest, Ruff, mypy and ESLint

Python, FastAPI, PostGIS, Docker Compose, React, TypeScript

View source on GitHub
Oct to Nov 2026

Agentic-GameAI

A modular framework where LLM agents play five games of rising difficulty, from FrozenLake to a four-snake Arena. One shared core handles the retrying LLM client, JSONL step logs and the Pygame viewer.

  • Sokoban planning with deadlock detection, retries from past failures and an optional BFS hint
  • Real-time Breakout agent: pixel-to-state extraction, per-frame reflexes, a background LLM for strategy
  • Snake Arena where LLM snakes face algorithmic opponents

Python, LLM APIs, Gymnasium, ALE, Pygame

View source on GitHub

Experience

Fiverr

Freelance, Unreal Engine (C++), Remote

  • Delivered 9 end-to-end projects for 7 international clients on tight deadlines, owning requirements, C++ and Blueprints implementation, and handoff.
  • Fixed crashes, logic errors and performance bottlenecks across 3 production projects, improving stability across devices.

Achievements

  • Top 10 of 2,000+ teams at Fintechstico (NSUT)
  • Top 50 of 2,700+ teams at Brainwave (GGSIPU)
  • Top 50 of 1,500+ teams at AzinHack (GGSIPU)
  • 300+ LeetCode problems solved, contest rating 1400+
  • i7 Lead at IoSC GGSIPU-EDC, leading a team that builds AI/ML and game projects

Skills

AI and ML
Deep learning, computer vision, RAG and agentic systems.
Web
FastAPI and Node.js back ends, React and TypeScript front ends.
DevOps
Docker, Kubernetes, GitHub Actions and AWS on Linux.

Contact

tusharsingh1346@gmail.com

GitHubLinkedInLeetCodeKaggleYouTube