Shreya Sengupta

Shreya Sengupta

ECE undergraduate at BITS Goa focused on building production-grade backend systems used by real users. Experienced in designing scalable workflows, distributed task processing, and ML-backed data pipelines, with hands-on exposure to taking systems from prototype to production under real-world constraints

Tech-Stack

• Programming Languages: Python, C, C++, HTML/CSS, JavaScript, familiar with Java, Verilog, Dart, Flutter.
• Machine Learning & Deep Learning: (Supervised Learning, Ensemble Methods), Random Forests, Federated Learning (basics, Flower); Scikit-learn, PyTorch, TensorFlow
• Data Analysis & Visualization: Pandas, NumPy, Matplotlib
• ML Systems & Deployment: Docker
• Tools & Platforms: Git, GitHub, Linux, Jupyter Notebook, Android Studio, Codeforces, Leetcode

In process of learning Data Structures and Algorithms

Projects

Campus Inventory Management Application

• Co-developed a production-grade inventory management system used by ~100 rotating users annually to manage AVU equipment across campus events and festivals.
• Solved a long-standing operational issue by replacing manual tracking with a Google-authenticated, centralized workflow, improving equipment traceability post-events.
• Application is actively used by hired department staff and will continue as a permanent operational system.
• Collaborated in a 20-member team, contributing across frontend, backend, and database components.

Distributed Task Queue System (Go, Redis Streams)
(Ongoing)

• Built a production-style distributed task queue supporting retries, delayed jobs, and dead letter queues.
• Leveraged Redis Streams consumer groups for reliable message delivery and automatic worker crash recovery.
• Implemented idempotent job execution and duplicate handling to ensure correctness under failures.
• Containerized services using Docker Compose to demonstrate horizontal scalability and fault tolerance.

Accelerometer Data Generator using Random Forests

• Designed and implemented a synthetic accelerometer data generation pipeline using Random Forest models.
• Simulated realistic sensor signal patterns for downstream machine learning tasks. Evaluated generated data using statistical analysis and visual inspection to validate distributional similarity.
  ECG Signal Denoising and Arrhythmia Classification
• Built an ECG analysis pipeline on the MIT-BIH Arrhythmia Database using digital filters and a Pan-Tompkins algorithm-based method for noise removal and R-peak detection.
• Engineered HRV and morphological features and trained a Random Forest model to classify Normal, PVC, PAC, and Bradycardia rhythms.

ECG Signal Denoising and Arrhythmia Classification

• Built an ECG analysis pipeline on the MIT-BIH Arrhythmia Database using digital filters and a Pan-Tompkins algorithm-based method for noise removal and R-peak detection.
• Engineered HRV and morphological features and trained a Random Forest model to classify Normal, PVC, PAC, and Bradycardia rhythms.

AI-Chatbot - RAG (using FastAPI)

• Built a fully local RAG-based AI chatbot using Streamlit, Ollama, LangChain, and ChromaDB.
• Implemented PDF document retrieval with embeddings, semantic search, streaming responses, and conversational memory using local LLMs such as Qwen3/Llama3.
• Completely open sourced