Hi, I'm NishankHi, I'm Nishank 👋

• Contributing to the AgiMo project, exploring Vision-Language Models and knowledge graphs for understanding urban streetscapes and street-design elements.

Tech Stack: Python, Vision-Language Models, Knowledge Graphs, Geospatial AI

• Developed and deployed a production AI-powered raw-material price modelling workflow for the Supplier Development Group covering ~300 feedstocks across ~5 years of data, combining FastAPI, RAG and ChromaDB for internal web/intranet content with an agentic LLM workflow and structured tool/function calling in Azure AI Foundry for live market information.
• Designed and automated a raw-material index forecasting pipeline for the Controlling group using monthly data from 2008 to 2023 across ~30 material indices, achieving ~14% MAPE and eliminating approximately 6 man-weeks of manual forecasting effort; presented the results in an internal technical talk.
• Built and maintained Continental's AI Knowledge Hub on SharePoint, consolidating AI tools, guidelines, Responsible AI, EU AI Act resources and AI news; reached 6,000+ views on the main page and ~10,000+ cumulative views across sub-pages, while developing AI use-case analyses and dashboards using Excel, Power BI and PowerPoint.
• Automated Business Partner Access workflows using Power Automate across 3 Business Partners and 4 business functions, and developed a structured framework for Citizen Code Compass to evaluate low-code automation solutions across 14 parameters, supporting technology and automation decisions.
• Contributed to SmartIdent, an industrial computer-vision initiative involving image classification and semantic segmentation of tire components, developing structured image datasets covering load-speed indices, tire markings, dimensions and other visual attributes to support downstream computer-vision model development.

Tech Stack: Python, RAG, LLMs, AI Agents, FastAPI, ChromaDB, Azure AI Foundry, Tool/Function Calling, Selenium, ARIMA, LightGBM, Power Automate, SharePoint, Power BI

• Developed an LSTM-based performance-monitoring system for the Antenna Control Servo System (ACSS) using ~5M time-series records across 30+ operational parameters, achieving MAE of approximately 0.06 for early failure detection.
• Analysed Antenna Control Unit (ACU) telemetry and historical operational data, identifying 12 degradation and failure patterns and deriving key parameters to characterize system behaviour and support predictive-maintenance decisions.
• Automated continuous system health assessment, reducing reliance on manual inspection and enabling earlier detection of degradation and failures; co-authored and presented "Performance Prediction of Antenna Control Servo System Based on LSTM Network" at IEEE SPACE 2024 Conference, organised by ISRO and DRDO.

Tech Stack: Python, TensorFlow, NumPy, Pandas, scikit-learn, LSTM, Time-Series Analysis, Anomaly Detection, Statistical Analysis

• Developed an automated computer-vision pipeline to analyse Typhifast Immunochromatographic Test (ICT) card images and automatically update diagnostic results in the Laboratory Information System (LIS), achieving 97% classification accuracy with ~2 to 3 ms per-image inference time, including preprocessing and model inference.
• Implemented segmentation, preprocessing, augmentation and deep-learning classification, evaluating architectures including InceptionNet, DenseNet121, MobileNetV2, VGG-19, EfficientNet and ResNet variants; used Docker and MLflow to support reproducible model deployment and experiment tracking.

Tech Stack: Python, TensorFlow, OpenCV, Docker, MLflow, Computer Vision, Deep Learning, Model Deployment

Performance Prediction of Antenna Control Servo System based on LSTM Network

Published and Presented in IEEE Space, Aerospace and Defence Conference 2024 (Scopus Indexed)

Research article focused on developing a predictive maintenance model through equipment performance monitoring for the Antenna Control Servo System (ACSS) using a stacked LSTM network. By employing a sliding window approach, the model processes sequential data from servo logs to predict the future current demands of the servo elevation motor during satellite passes. The study demonstrates the model's effectiveness, achieving a mean absolute error of 0.06 in predicting elevation current, thereby enabling proactive detection of equipment degradation and failures.

Optimizing Traffic Management Through Density-Driven Dynamic Traffic Signaling and Emergency Vehicle Prioritization Using Audio and Video

Published and Presented in the 7th International Conference on Innovative Computing and Communication (Scopus Indexed)

Research article focused on addressing the challenges of increasing traffic congestion and its adverse effects on daily life and emergency services. The paper presents a hybrid system that employs deep learning models for dynamic traffic signaling and real-time detection of emergency vehicles. Vehicle Detection uses Single Shot Detector (SSD) and Emergency Vehicle Detection combines audio and video inputs to confirm the presence of an emergency vehicle.

Innovating Drug Design for Alzheimer's Disease via Reinforcement Learning for Enhanced Molecular Generation

Published and Presented in the 4th International Conference on Innovations in Computational Intelligence and Computer Vision (Scopus Indexed)

Research article introducing an innovative drug design approach for Alzheimer's disease, leveraging AI techniques such as a random forest predictor, stacked GRU architecture, and reinforcement learning. The study integrates machine-learning algorithms to assess and rank compounds based on target characteristics, offering a promising avenue for efficient drug discovery in the context of Alzheimer's disease.