Key Projects

AYUSHYA Your Chronic Care Companion

Duration: January 2025 - Present | Technologies: Python, PyTorch, LLM, Agentic AI

See [Details]

ZenifyPro: Personalized Persuasion Engine for Health Care

Duration: January 2024 - December 2024 | Technologies: Python, PyTorch, Reinforcement Learning

• Developing an end-to-end system for generating persuasive notifications for Sleeping Disorder patients.
• Utilized LLM, Langchain for notification generation.
• Used prompt engineering, in-context learning, chain-of-thought.
• Designed algorithms for on-board persuasion strategy adaptation.

SCOPE-API: Skill Extraction from Job Description and CV

Duration: August 2023 - March 2024 | Technologies: Python, PyTorch, Flask, SpaCy, Hugging Face

• Built a Skill extraction API for a talent match engine.
• Implemented pre-trained transformer models using Hugging Face.
• The API is currently being deployed on an Azure-based cloud platform.

Achievements: The model outperforms GPT-4. It is used internally by the talent search team. A patent has been filed.

LUCIFER: Fake News Verification and Explanation Generation

Duration: August 2022 - July 2023 | Technologies: Python, PyTorch, Hugging Face, Meta learning

• Built a domain-agnostic real-time fake news verification and explanation system.
• Implemented pre-trained transformer models using Hugging Face.
• Implemented meta-curriculum learning for generalization across domains and few-shot adaptation.
• The API is currently being deployed on an Azure-based cloud platform.

EXCLAIM: Meme role labelling and explanation generation

Duration: August 2021 - July 2022 | Technologies: Python, PyTorch, OCR, Hugging Face, OpenCV

• Built a multimodal meme explanation system to detect and label Hero, Villain, and Victim roles in harmful memes.
• Generate explanations for the actions of each role within the meme.
• Implemented a pipeline process using OpenCV, OCR, natural language understanding, and caption generation from images, validated with OCR and pre-trained transformer models from Hugging Face for explanation generation.

Other Projects (Industry-Academia collaboration)

Duration: March 2020 - July 2022 | Technologies: Python, PyTorch, Hugging Face

• Hate Speech detection, Normalization and Diffusion.
• Document level Machine Translation.
• Lecture Video translation.

Achievement: One A and three A-level research publications.

Postdoc Projects

Duration: February 2017 - February 2020 | Technologies: Python, PyTorch, Transformers

• MMPE: Multi-modal and Language Technology Based Post-Editing Support for Machine Translation" sponsored by the German Research Foundation (DFG) (Saarland University, Germany).
• DEEPLE: Deep Learning for Language Technology End-to-End Applications (German Research Foundation for Artificial Intelligence (DFKI), SB, Germany).
• APE++: Neural Automatic Post-Editing (German Research Foundation for Artificial Intelligence (DFKI), SB, Germany).

Achievements: Four A* research publications for MMPE, and one A and two A-level research publications for DEEPLE.