Robotics & AI Engineering student at the University of Engineering and Management (UEM) Kolkata with hands-on experience developing autonomous systems, advanced computer vision integration, and large language model (LLM) workflows. Proven capability in building end-to-end physical quadruped locomotion platforms, deploying real-time ROS 2 navigation stacks with YOLOv8 perception libraries, and building multi-agent RAG-based systems. Recognized open-source contributor to scikit-learn and academic conference presenter.
Technical Expertise
Core Programming Languages
Python, C++, C, SQL
Robotics & Frameworks
ROS 2 Jazzy, Gazebo, Nav2, SLAM Toolbox, RViz, PID Control, A* / DWA Path Planning
Machine & Deep Learning
PyTorch, TensorFlow, Scikit-learn, Keras, YOLOv8 (Ultralytics), OpenCV
LLM, RAG & Agentic Systems
LangGraph, LlamaIndex, RAG (ChromaDB / FAISS), Ollama, Streamlit
Hardware Integration
Arduino, ESP32, Raspberry Pi, LIDAR, IMU, Ultrasonic / IR Sensors, Servo PWM, Motor Drivers
Systems, Tools & Data Science
Linux (Ubuntu), Docker, Git/GitHub, Power BI, Pandas, NumPy, Matplotlib, Seaborn, Jupyter
Featured Projects
ArduinoESP32C++12× MG996R ServosInverse KinematicsESP-NOWServo PWM
- Designed and fabricated a fully articulated quadruped robotic system with 3 Degrees of Freedom (DOF) per leg, utilizing custom 3D-printed joint brackets and a rugged structural aluminum chassis.
- Implemented a bio-inspired gait pattern via iterative servo calibration for multi-terrain locomotion; leveraged a symmetric leg pairing mechanism to guarantee robust static stability.
- Developed real-time, low-latency wireless control firmware using the ESP-NOW protocol on dual ESP32 chips, driving sub-5 ms command latency execution for high-fidelity servo PWM generation.
- Peer-reviewed paper accepted and formally presented at the TSSC 2026 conference.
ROS 2 HumblePythonNav2SLAM ToolboxGazeboYOLOv8
- Simulated a high-fidelity differential-drive robot inside Gazebo integrated with sensor-fusion payloads containing LIDAR, IMU, and custom wheel odometry feedback.
- Implemented simultaneous localization and mapping (SLAM) using the SLAM Toolbox, generating high-accuracy localized occupancy grid maps at a resolution of less than 3 cm.
- Deployed the full ROS 2 Nav2 stack configured with an A* global planner, a Dynamic Window Approach (DWB) local planner, and custom behavior trees to handle dynamic recovery loops, reaching a 92% goal success rate over 50+ runs.
- Integrated an asynchronous YOLOv8 vision pipeline node as a native ROS 2 subscriber to stream dynamic obstacles straight into the local costmap layer for real-time local cost adjustments.
PythonOllamaLangGraphLlamaIndexVector DatabasesStreamlit
- Developed an end-to-end academic paper summarizer powered by a multi-agent orchestration layer, automating high-fidelity document retrieval, synthesis, and granular deep analytical reasoning via RAG.
- Implemented a LangGraph-based agentic workflow utilizing state-driven conditional routing and LlamaIndex to execute precise context-aware multi-document reasoning over complex research corpora.
PythonOllamaAST ParsingStatic AnalysisFastAPICI/CD Pipeline
- Engineered a static analysis security tool that constructs and traverses Abstract Syntax Trees (ASTs) to detect code-level structural vulnerabilities before passing contextual code blocks to LLMs for remediation.
- Developed an automated patch-generation loop combining precise deterministic rules with generative AI reasoning to auto-generate secure, context-aware refactored code patches.
- Designed a high-throughput FastAPI backend to handle code stream parsing and serve real-time vulnerability telemetry, structured for seamless integration into DevOps CI/CD pipelines.
Professional Experience
AI/ML Intern (Remote)
- Built an advanced customer segmentation pipeline executing unsupervised K-Means clustering over a dataset of 5,000+ consumer records, identifying 4 distinct customer archetypes to back targeted digital marketing strategies.
- Engineered an extensive end-to-end data preprocessing architecture using Pandas and NumPy, handling missing-value imputations, algorithmic outlier removal, and statistical feature engineering to bump downstream model accuracy by 12%.
- Designed and deployed interactive Power BI enterprise dashboards to monitor behavioral patterns; pitched analytical findings directly to an audience of over 50 corporate and technical attendees.
Publications & Peer-Reviewed Presentations
"Bluetooth-Enabled 3D Printed Quadruped Spider Robot on Arduino Nano: A Socio-Technical Study Bridging Indian Knowledge Frameworks and Community Empowerment"
S. Banerjee, A. Ghosh, S. Bhadra, R. Saha, P. Das, and A. Mukherjee.
Presented at the 1st International Conference on Transformative Social Sciences (TSSC 2026): Bharat's Wisdom for Global Solutions, UEM Kolkata, February 2026.
Open Source Contributions & Achievements
- 🔹 Open Source Contributor @ scikit-learn: Authored and merged PR #32677 optimizing performance within the feature-scaling preprocessing pipeline. The algorithmic enhancement directly accelerated overall pipeline performance by 15%, fully approved and integrated by the core scikit-learn maintainer team.
- 🔹 Conference Presenter @ TSSC 2026: Awarded official certificate and accepted paper status for novel hardware architecture demonstrations regarding 3D-printed multi-terrain quadruped platforms at UEM.
Education
University of Engineering and Management (UEM)
B.Tech in Robotics and Artificial Intelligence Engineering
2024 – 2028 (Expected)
Kolkata, West Bengal
Core Track: Data Structures & Algorithms, Machine Learning, ROS 2, Computer Vision, Control Systems, Embedded Systems, Path Planning.
Professional Certifications
• Machine Learning Specialization | Stanford Univ. & DeepLearning.AI
• Computer Vision Engineer Professional Certificate | MathWorks
• Data Analytics Professional Certificate | DeepLearning.AI