Robotics & AI Engineer

Arpan
Mukherjee.
Builder.

B.Tech in Robotics & AI at UEM Kolkata. I build autonomous systems, train deep learning models, and ship things that move in the real world.

servo_1 servo_2 servo_3 S_R1 S_R2 S_R3 ESP32 / Arduino 18x MG996R span: 300mm SYS_NOMINAL
8.0/10
CGPA — UEM Kolkata
5+
Projects Shipped
1st
Open Source Contributor
Top 2%
Kaggle (15k participants)

Robots, Models, and Code

I'm a Robotics & AI Engineering student at the University of Engineering and Management, Kolkata (2024–2028), ranked in the top 15% of my cohort. I sit at the intersection of autonomous systems and machine intelligence.

My work spans the full stack — from writing CUDA kernels and SLAM pipelines to building multi-agent LLM systems and RAG applications. I care deeply about systems that work reliably in the real world, not just in notebooks.

I completed an AI/ML internship with IBM SkillsBuild (via Edunet Foundation), contributed a merged PR to scikit-learn, won 1st place at UEM's robotics challenge, and publish technical writing on Medium reaching 500+ readers.

🤖
Robotics & Embedded ROS2, Arduino, ESP32, Servo Control, SLAM
🧠
Machine Learning & AI PyTorch, YOLOv8, LangChain, RAG, LLMs
⚙️
Systems Engineering CUDA, Docker, Linux, Git, Distributed Training
📊
Data & Analytics Pandas, Power BI, Scikit-learn, Kaggle Silver

What I Build With

Languages
Python C++ C SQL HTML/CSS CUDA Bash
ML & AI Frameworks
PyTorch YOLOv8 TensorFlow Scikit-learn Keras LangChain HuggingFace OpenCV
LLM & Agents
RAG ChromaDB FAISS LangGraph Gemini API OpenAI API Streamlit
Robotics
ROS2 Humble Gazebo Nav2 SLAM Toolbox RViz PID Control A* Planning Arduino ESP32
Systems & Tools
Linux / Ubuntu Docker Git / GitHub ONNX TorchScript Power BI VS Code
Sensors & Hardware
LIDAR IMU Ultrasonic IR Sensors Camera Motor Drivers Raspberry Pi Servo PWM

Projects that matter

02
Autonomous Navigation — ROS2
ROS2 Humble Nav2 SLAM Gazebo Python In Build

Full autonomous navigation stack — SLAM mapping, A* global planning, DWB local obstacle avoidance, and YOLOv8 vision integration. 92% goal success across 50+ sim trials.

  • SLAM Toolbox generating occupancy grids at <3cm resolution over 20m×20m environments
  • Custom Python navigator node with real-time distance feedback and failure recovery
  • YOLOv8 obstacle detector publishing to costmap as a ROS2 node
  • Behaviour tree with spin/backup/wait recovery for stuck states
03
Multi-Agent AI Research Assistant
Gemini API RAG LangGraph Python In Build

An automated Paper Summarizer and research assistant. Utilises a multi-agent architecture and RAG system powered by the Google Gemini API to retrieve, analyze, and synthesize academic papers efficiently.

  • Built document ingestion pipelines chunking PDFs for vector storage
  • Implemented LangGraph for orchestrating specialized search and summary agents
  • Integrated the Gemini API to handle high-context summarization and complex reasoning
  • Reduced manual research reading time by extracting key methodologies and findings
04
Distributed ML Training
PyTorch CUDA Docker Python In Build

A scalable framework utilizing PyTorch's DistributedDataParallel to train deep learning models across multiple GPU nodes, significantly reducing training time for large datasets.

  • Implemented cross-node gradient synchronization strategies
  • Containerized the training environment using Docker for seamless deployment
  • Optimized data loading bottlenecks to ensure high GPU utilization
  • Monitored training telemetry across cluster configurations
05
Code Vulnerability Scanner + LLM
LLMs AST Parsing Security Python In Build

An AI-augmented security tool that parses abstract syntax trees (AST) of codebases and leverages LLMs to identify potential vulnerabilities, suggesting secure refactors directly to developers.

  • Combined traditional static analysis techniques with LLM semantic reasoning
  • Generated automated and contextual patch suggestions for common CWEs
  • Built integrations to run scans automatically within CI/CD pipelines
  • Tuned prompts to reduce false positives in standard library usage
06
Federated Learning + Privacy ML
PyTorch Federated Learning Python In Build

A privacy-preserving machine learning framework simulating distributed edge devices. Models are trained locally on device partitions, and only aggregated weights are shared securely with the central server.

  • Implemented the Federated Averaging (FedAvg) algorithm from scratch
  • Simulated network latency, node dropout, and non-IID data distributions
  • Ensured differential privacy bounds by applying noise to gradient updates
  • Benchmarked model convergence rates against centralized baseline training

Where I've Worked

Jan 2026 — Feb 2026
Edunet Foundation × IBM SkillsBuild
AI/ML Intern — Remote
  • Built customer segmentation model using K-Means clustering on 5,000+ records, identifying 4 distinct groups for targeted marketing campaigns
  • Developed data preprocessing pipeline handling missing values, outliers, and feature engineering with Pandas — improved model accuracy by 12%
  • Created interactive Power BI dashboards visualising customer insights; presented findings to cohort of 50+ participants
  • Tech Stack: Python, Scikit-learn, Pandas, NumPy, Power BI

Achievements & Contributions

🔧
Merged PR — scikit-learn (Open Source)
Contributed optimisation to feature scaling preprocessing — merged into main branch, improving performance by 15%.
✍️
Technical Blogger — 500+ Readers on Medium
Published 5+ in-depth articles on ROS2, deep learning, and robotics fundamentals reaching 500+ total views.
Deep Learning Specialization
DeepLearning.AI — Coursera
Mar 2026
ROS2 for Beginners: Basics & Navigation
The Construct
Feb 2026
Data Analytics Professional Certificate
DeepLearning.AI
Jan 2026
Oracle Cloud Infrastructure AI Foundations
Oracle
Nov 2025

Let's build something real.

Open to internships, research collaborations, and interesting problems in robotics, autonomous systems, and applied AI. Based in Kolkata — available remotely.

Send a message →