Hi, I'm Tanmay Kumar Sahu

|

Backend Developer @ SkyIT  ·  MS Computer Science @ NYU  ·  Brooklyn, NY

01

About

I'm an NYU MS Computer Science student and backend developer building AI/ML systems, production APIs, and data infrastructure. My work spans computer vision research, edge-native AI wearables, RAG and multi-agent pipelines, and scalable backend services for real-world product teams.

Comfortable moving between theory, implementation, and deployment: from training segmentation models on A100 GPUs to building Django REST, Express, GraphQL, Docker, and database-backed systems across PostgreSQL, MySQL, MongoDB, and Azure workflows.

Published in IEEE & Springer
Backend APIs · Full Stack Systems
AI/ML · Edge Inference · RAG
NVIDIA · Intuit · IBM Hackathon Winner
Location Brooklyn, NY
GitHub 7dracoder ↗
LinkedIn sahu-tanmay ↗
02

Skills

Programming

PythonTypeScriptJavaScriptRJavaSQLSwiftKotlinMATLAB

ML Frameworks

PyTorchTensorFlowScikit-learnJAXKerasXGBoostLightGBMOpenCVHugging Face

Deep Learning & AI

LLM Fine-tuningRAGMulti-agent PipelinesTool-use / Function CallingPrompt EngineeringMultimodal AIComputer VisionU-NetYOLOv8

MLOps & Infrastructure

GCP (Vertex AI, Cloud Run)AWSAzureDockerRedisWebRTCMLflowPostgreSQLMySQLMongoDBFastAPI

Data Engineering

HadoopApache SparkPandasNumPyStreamlitPower BID3.jsPlotly

Web & Full Stack

Next.jsReactDjango REST FrameworkExpress.jsFlaskNode.jsGraphQL APIsREST APIsJWTSupabaseAppwrite
03

Selected Work

01
Winner

MetaNemo: Edge-Native AI Wearable

Edge-native AI wearable on NVIDIA DGX Spark GB10 running all inference locally, no cloud APIs. Won Most Secure Project at NVIDIA Spark Hackathon (Pensar & NVIDIA).

PythonNode.jsFastAPINVIDIA NeMoPostgreSQLPostGISDockerSwift
02

Semantic Segmentation: Autonomous Lane Detection

U-Net encoder-decoder with ResNet-34 backbone on CULane (95k+ images, A100 GPU). Achieved 97.1% mean IoU, 22-32% above SOTA, at 67 FPS, 2.2× faster than the 30 FPS real-time requirement.

PythonPyTorchU-NetResNet-34CULaneA100 GPU
03

Pixel Glass: Multimodal AI for Smart Glasses

AI assistant for Meta Ray-Ban Smart Glasses answering real-world queries in under 5 seconds using Google Gemini Live 3.1 Flash Lite. Multi-agent backend with RAG pipeline on GCP, native iOS & Android apps.

PythonGoogle Gemini LiveGCPRAGWebRTCSwiftKotlin
04
IEEE

Real-Time Object Detection: Model Benchmarking

Benchmarked YOLOv8, Faster R-CNN, MobileNetV2, and EfficientDet D4 on real video streams. Published in IEEE. YOLOv8 outperformed all others across FPS, accuracy, precision, and confidence.

PythonYOLOv8Faster R-CNNOpenCVPyTorch
05
Springer

Video Dehazing: Unsupervised Deep Learning

Unsupervised image restoration using Dark Channel Prior Loss (arXiv:1812.07051). No paired training data required. Extended to video frames for real-time dehazing in autonomous vision. Published in Springer.

PythonTensorFlowCNNDark Channel PriorOpenCV
06

VisionCare AI: Healthcare Patient Management

Full-stack patient management platform with OTP-based passkey auth, Zod validation, admin dashboard with appointment status management, and real-time error monitoring via Sentry.

TypeScriptNext.jsAppwriteTailwind CSSSentry
07

SkyBook: Airline Flight Search & Booking

Full-stack Flask + PostgreSQL app with normalized relational schema (6 tables), real-time seat availability via complex SQL JOINs, idempotent seeding, and production-ready env-based config.

PythonFlaskPostgreSQLpsycopg2SQL
08

Lemontree Pantryhub: Food Security Analytics

Food security platform aggregating 14,178+ real-time pantry resources with federal demographic data. ML scoring pipeline for optimal food bank placement. Voronoi coverage maps, heatmaps, AI assistant. Morgan Stanley Hackathon.

ReactPythonSupabaseMLD3.jsVercel
09

YTRecipe Agent: Multi-Agent LLM Pipeline

Multi-agent pipeline extracting YouTube transcripts, parsing recipe steps, normalizing ingredients, fetching nutritional data, and generating a grocery cart from a single URL. Redis-backed caching and state.

PythonLLMMulti-AgentRedisStreamlit
10

Critics vs. Consumers: Video Game Analytics

Interactive dashboard analyzing 2,000+ video game titles from 1985–2016. Cross-filtering across genre, platform, region, and publisher. Explores critic score vs. consumer sales divergence.

JavaScriptPlotly.jsD3.jsEDA
11

Artisan: AI Content & Visual Design Platform

AI-driven content and visual creation project focused on faster design workflows and smarter creative asset generation.

HTMLCSSJavaScriptAIDesign Tools
12

CodeLab GenAI

Python generative AI code lab exploring model prompting, AI workflows, and practical GenAI experimentation.

PythonGenerative AIPromptingAI Workflows
13

Forecasting Smart City Traffic Patterns

Jupyter Notebook project analyzing historical traffic patterns to identify trends and recommend optimized smart-city traffic management and infrastructure planning.

Jupyter NotebookForecastingData AnalysisSmart City
14

Light Chatty Bot

Simple Python chatbot project practicing conversational flow, input handling, and lightweight command-line interaction.

PythonChatbotCLIConversation Logic
15

Readability Score Master

Java text analysis tool calculating readability difficulty and estimated reader age using ARI, Flesch-Kincaid, SMOG, and Coleman-Liau metrics.

JavaText AnalysisRegexReadability Metrics
16

Battleship

Pure Java recreation of the Battleship board game with grid state, turn logic, and object-oriented gameplay structure.

JavaOOPGame LogicCLI
04

Experience & Education

2026

SkyIT - GBCS Group

Backend Developer

June 2026 – Present · New York, USA

  • Developing and optimizing backend services for fleet management applications using Django REST Framework, Express.js, REST APIs, GraphQL APIs, Git, and Docker
  • Building scalable server-side infrastructure aligned with product, frontend, and DevOps requirements
  • Supporting database management, testing, deployment, and performance troubleshooting across PostgreSQL, MySQL, MongoDB, and Microsoft Azure workflows
2023

UniConverge Technologies

Data Science & Machine Learning Engineer

May 2023 – July 2023 · Noida, India

  • Built and deployed Decision and Extra Tree Classifier models for smart city optimization using Python, TensorFlow, and Scikit-learn, improving planning accuracy by ~15–20%
  • Cleaned, merged, and analyzed large traffic and weather datasets with SQL, Pandas, and Excel
  • Engineered temporal features and evaluated performance with MAE; integrated real-time adaptive learning
2023

Stanford CodeInPlace

Section Leader & Tutor

2023 · Remote

  • Taught Python fundamentals, data structures, file operations, OOP, to 10 students
  • Mentored students through their debut projects
2022

Amity AI Club

Team Lead

2022 – 2024 · India

  • Led the club and organised webinars, inter-university hackathons, and workshops on AI (NLP, Bridging AI to the Edge)
  • Mentored members on TensorFlow, PyTorch, and Python libraries; guided individual project development
2025

New York University

Master of Science in Computer Science

Sep 2025 – May 2027 · New York, USA · GPA: 3.8

  • Coursework: Big Data, Programming in Data Science, Design and Analysis of Algorithms, Machine Learning, Information Visualization, DBMS
2020

Amity University

Bachelor of Technology in Computer Science & Engineering

Sep 2020 – Nov 2024 · UP, India · CGPA: 8.0

  • Focus: AI/ML, Data Structures, Algorithms
  • Published 2 peer-reviewed research papers (IEEE, Springer)
05

Research

IEEE · 2024

Real Time Video Object Detection using Deep Learning

Kunal Sharma, Tanmay Kumar Sahu, Prateek Bhatt, Aniket Kumar Singh, Madhulika Bhatia · Amity School of Engineering

Benchmarked YOLOv8, Faster R-CNN, SSD MobileNetV2, and EfficientDet D4 on real video streams. YOLOv8 outperformed all others across FPS, accuracy, precision, and confidence. Dataset included security camera footage and self-recorded highway/campus videos.

Springer · 2024

Real Time Video Dehazing & Object Detection using Deep Learning

Juhi Singh, Tanmay Kumar Sahu, Kunal Sharma, Prateek Bhatt, Aniket Kumar Singh, Sapna Gambhir (George Mason University)

Integrated pipeline combining Dark Channel Prior dehazing and MAP-Net video dehazing with YOLOv8 detection. Evaluated on REVIDE dataset using IoU, SSIM, and PSNR. Applications in autonomous driving, surveillance, and environmental monitoring.

06

Let's Connect

ts5789@nyu.edu

Open to AI/ML Engineering, Backend Engineering, Data Science, and Research roles

Brooklyn, NY · Available for full-time & internship opportunities