Beyond Factual QA: Mentorship-Oriented Question Answering over Long-Form Multilingual Content (MentorQA)
Accepted: EMNLP 2026 — Budapest, Hungary
This work introduces MentorQA, a multilingual benchmark and evaluation framework for mentorship-oriented question answering over long-form video content. The dataset contains nearly 9,000 QA pairs from approximately 180 hours of content across four languages. We compare Single-Agent, Dual-Agent, RAG, and Multi-Agent architectures and evaluate dimensions beyond factual correctness, including clarity, relevance, coherence, and mentorship quality.
When Cultures Meet: Multicultural Text-to-Image Generation (MosAIG)
Published: ACL 2026 — San Diego, California
This work introduces multicultural text-to-image generation as a new task and presents a 9,000-image multilingual benchmark spanning five countries, five languages, multiple demographic groups, and historical landmarks. We evaluate alignment, image quality, aesthetics, cultural knowledge, and fairness, and explore MosAIG, a multi-agent framework using culturally specialized LLM agents to improve prompt composition and cultural grounding.
When Cultures Move: Measuring and Improving Multicultural Text-to-Video Generation (MAVEN)
Accepted: EMNLP 2026 : NLP4PositiveImpact Workshop — Budapest, Hungary
This work studies multicultural text-to-video generation and introduces MAVEN, a multi-agent prompt refinement framework for improving cultural fidelity in mono-cultural and cross-cultural video generation. The study evaluates 243 culturally grounded scenarios and 972 generated videos using CLIP, VLM-as-a-Judge, cultural relevance, visual quality, and temporal consistency.
ECG Classification Using Machine Learning on Wave Samples for the Indian Population
Published: 2023 – IEEE International Conference on Advancement in Computation & Computer Technologies (InCACCT)
This study proposes machine learning models for ECG rhythm classification tailored to the Indian population, using only 3 leads instead of the conventional 12, thereby reducing cost and complexity. The models classify cardiac rhythms as normal or abnormal directly from ECG wave samples. Multiple algorithms are compared for accuracy, using a dataset collected from hospitals in Nagpur city, making the work both practical and population-specific.
Design and Implementation of a Dynamic Traffic Signal System with Digital Circuit and IoT Integration for Efficient Traffic Management
Published: 2023 – IEEE IIT-Delhi Top Conference, 14th International Conference on Computing Communication and Networking Technologies (ICCCNT)
This research proposes a dynamic, sensor-based traffic signal system to overcome inefficiencies of static traffic management in India. The system adapts signal timing to real-time traffic density and incorporates an IoT-based emergency override for ambulances and VIP movement. A decoder-driven digital circuit, implemented in Verilog and simulated with a Python GUI, enables multiple active lanes without interference. Results show improved traffic flow, reduced congestion, and faster emergency response, offering a scalable solution for modern cities.
Performance Analysis of the YOLOv5 Algorithm for American Sign Language Detection
Published: SSRN Journal, 2022
This paper analyzes the YOLOv5 algorithm for American Sign Language (ASL) detection, evaluating its performance across TensorFlow and PyTorch frameworks on devices such as Intel processors and Raspberry Pi. The study identifies alphabets with higher false positive/negative rates and compares model accuracy under different setups. Results indicate that Raspberry Pi with TensorFlow fp16 at 320 image size and weight is the most efficient configuration for practical deployment.
Point of Care Device for Measurement of Vital Parameters
Published: Springer, 2023 – Smart Trends in Computation and Communication (SmartCom 2023 International Conference)
This research proposes a solution to improve healthcare access in remote areas by developing a portable, non-invasive device. This device is capable of measuring vital health parameters like body temperature, ECG, and PPG, and then calculating heart rate and blood pressure. The study focuses on creating a compact, low-cost prototype to enable early diagnosis and reduce the need for patients to travel to city hospitals, with further recommendations for improving accuracy. (Patent Granted)
