Rohit Prajapat
Data Scientist from IIT Madras specializing in Computer Vision and Natural Language Processing. Proven track record of deploying real-time perception systems (object detection, depth estimation) and building complex, multimodal data pipelines. Passionate about LLMs (BERT, LLaMA) to solve challenges.
Experience
Software Development Intern
Sahay Ai
Bengaluru
Engineered and shipped a Model Context Protocol (MCP) server, building scalable infrastructure to interface advanced AI models with local environments. Developed an integrated Chrome extension, focusing on high-performance system design and creating seamless user/AI interaction loops.
Education
Indian Institute of Technology Madras
BSc
Skills
NLPLLaMABERTLangchainLangGraphPythonPyTorchCUDADeep Learning ArchitecturesAPI IntegrationGitGoogle Cloud Platform (GCP)PandasNumpyScikit-learnSparkPostgresqlChromaDBPower BITableauData Annotation PipelinesDatabase Architecture (SQL)
PROJECTS
AI Perception Project | Navigation Assistant for the Visually Impaired
Core focus: Real-time spatial perception, Computer Vision, Latency Optimization
Developed an offline, real-time computer vision system for visually impaired navigation using YOLOv8 Nano and lightweight monocular depth estimation on monocular RGB data. Preprocessed the NYU Depth V2 dataset with custom invalid depth masking, resolution normalization, and 0–3m depth clipping to optimize models for near-field obstacle avoidance.
Operationalized models for edge deployment by applying INT8 quantization and ONNX/TFLite compilation, achieving 7+ FPS inference without cloud compute dependencies. Engineered a deterministic spatial reasoning algorithm to translate raw bounding boxes and relative depth maps into immediate, rule-based directional safety commands.
System Threat Forecaster
Scikit-Learn, Pandas, Numpy, Matplotlib, Seaborn
Developed a binary classification model to predict the probability of malware infection on a system using telemetry and system log data. Performed extensive Exploratory Data Analysis (EDA) and feature engineering to handle class imbalance and high-dimensional categorical features. Optimized model performance using XGBoost, successfully surpassing the target accuracy criteria of 0.6325.
Vehicle Parking Management System
Flask, Vue.js, Redis, SQLite
Architected a multi-user parking management platform with Role-Based Access Control (RBAC) to securely separate admin and user functionalities. Developed a RESTful API using Flask, integrating Redis for high-performance caching to handle real-time slot availability checks. Implemented automated session management using Flask-Security, ensuring accurate billing based on parking Duration.
CERTIFICATIONS & ACHIEVEMENTS
Aspire Leaders Program Scholar
Selected for the competitive leadership program founded by Harvard faculty, focusing on cross-cultural collaboration and social impact strategies.
Winner, ETHIndia 2024 (BuidlGuidl Track)
Built 'OpinionX', a decentralized AI-powered opinion trading market.