DEEPTHI SUDHARSAN

Education

B.Tech, Artificial Intelligence 2019 — 2023

Amrita Vishwa Vidyapeetham, Coimbatore

CGPA: 9.13

Experience

Data Scientist Dec 2025 — Jan 2025

InvestCloud

• Worked as part of the Data, AI, and Analytics team contributing to their Digital Wealth product by evaluating and building a multi-agent intelligent assistant for wealth management advisors using LangGraph and GPT-based models.

Independent Researcher Jul 2025 — Dec 2025

In collaboration with Microsoft Research

• Advanced projects initiated during MSR fellowship through to paper submission at top-tier conferences and workshops (eg. AACL, AAAI, MMFood, etc.)
• Conducting continued research on AI agentic, culturally nuanced story generation to support financial knowledge acquisition among rural adults, while investigating community-centric evaluation frameworks for assessing cultural representation in generative AI systems.
• Collaborating with academic and industry researchers on dataset curation, benchmarking, and culturally nuanced storytelling methods for underrepresented contexts in Generative AI.

Research Fellow Jul 2023 — Jul 2025

Microsoft Research India

• Kahani — Led project Kahani, focused on building an AI-based multi-agent system for generating culturally-adaptive visual stories using GPT-4o and FLUX.1 and agentic frameworks like PydanticAI (Advisors: Dr. Kalika Bali and Dr. Mohit Jain).
• Inclusive Representation in T2I — Developed innovative evaluation metrics for long-tail concepts and under-represented communities in models such as Stable Diffusion 3 and DALL-E3 (Advisor: Dr. Cecily Morrison).
• Endangered Tribal Languages Documentation — Led collection and modeling of PoCs for various use cases, including fine-tuned RAG systems for traditional recipes in endangered tribal languages.
• Updesh — Helped build a large-scale synthetic dataset designed to advance post-training of LLMs for Indic languages using MediaWiki API, BingSearch API, and GPT-4o. (Advisors: Dr. Kalika Bali and Dr. Sunayana Sitaram).
• HyWay — Enhanced hybrid engagement by building privacy-preserving agents for summarization/recap tasks, deployed at internal events attended by Satya Nadella and Peter Lee.
• Azure Maps Creator Tool — Collaborated with Azure Maps team to build an AI-powered automation pipeline using GPT-4V with geospatial data and CAD files.

Technology Consulting Intern — Emerging Technologies Jan 2023 — Jul 2023

PwC India

• Delivered Proof-of-Concepts (from model development to model serving) for top clients, including automated weld defect detection using Deep Neural Nets and ANPR-based vehicle monitoring.

Machine Learning Intern Oct 2021 — Dec 2021

The Machine Learning Company

• Built end-to-end Machine Learning solutions to analyze Near-Infrared Spectroscopy measurements for ensuring consistent content uniformity of Active Pharmaceutical Ingredients using Linear Regression, SVM, Random Forests, and LightGBM.

Technical Skills

Publications

COMPASS 2025

KAHANI: Culturally-Nuanced Visual Storytelling Pipeline for Non-Western Cultures

Hamna, Deepthi Sudharsan, Agrima Seth, Ritvik Budhiraja, Kalika Bali, et al.

Developed Kahani, a pipeline utilizing GPT-4 Turbo and SDXL with CoT prompting to outperform ChatGPT-4 in cultural relevance.

View Paper

Preprint

The Role of Synthetic Data in Multilingual, Multicultural AI Systems: Lessons from Indic languages

Pranjal A Chitale, Varun Gumma, Sanchit Ahuja, Deepthi Sudharsan, Sunayana Sitaram, et al.

Presents Updesh, a synthetic instruction-tuning dataset for Indian languages built via a culturally grounded, bottom-up approach.

View Paper

Under Review

Towards a Community-Centric Approach to Measuring Cultural Representation in AI Image Generation

Nari Johnson, Deepthi Sudharsan, Hamna, et al.

Argues that AI measurement often ignores impacted communities, using case studies to show how lived experience shapes more meaningful metrics.

Preprint

To Make Text-to-Image Models that Work for Marginalized Communities, We Need New Measurement Practices for the Long Tail

Nari Johnson, Hamna, Deepthi Sudharsan, et al.

Highlights how current T2I metrics fail the long tail of low-resource concepts and calls for community-centered methods.

View Paper

DA&CI 2022

Coconut Tree Detection using Deep Learning Models

Deepthi Sudharsan, K. Harish, U. Asmitha, et al.

Optimized object detection models using IceVision framework for coconut tree detection and segmentation.

View Paper

ISCMM 2021

Brain Disorder Detection Using MRI Data

Deepthi Sudharsan, S Isha Indhu, Kavya S Kumar, et al.

Developed ML/DL models for diagnosis of Schizophrenia and Alzheimer's using MRI features.

View Paper

FIRE 2021

Rhetorical Role Labelling of Legal Case Documents

Deepthi Sudharsan, U Asmitha, B Premjith, KP Soman

Implemented DistilRoBERTa based sentence embeddings for rhetorical role assignment in legal documents.

View Paper

Awards & Honors

• Presented Kahani at BeyondCHI 2025, IndiaHCI 2025 and Cohere4AI Monthly Gathering.
• Selected for McKinsey Next Generation Women Leaders (APAC 2025) program among 300 globally.
• Advancing Inclusion Scholar at Grace Hopper Celebrations India (2024, 2025).
• Social Impact Award — Microsoft Research EMEA Hackathon 2024 for Kisaan Rafiki.
• Winner — 2024 GenAI InnovateHer Datathon for empowering visually impaired users.
• Philately — Multiple Competitive awards at Regional, National, and International Levels.

Certifications & Achievements

• Certified Scrum Product Owner (CSPO), Scrum Alliance.
• Google Developers ML Bootcamp Apprentice (Cohort-2).
• Founder & Head — First AI Club at Amrita Vishwa Vidyapeetham.
• Amazon ML Challenge 2021 — 111th place out of 3,600+ teams.
• 11th place — DTU-BITGRIT Viral Tweets Prediction Hackathon.