StoryForge AI Studio: Multilingual Story-to-Video Pipeline
End-to-end generative pipeline that turns keywords into narrated videos — story, scene images, audio, and final MP4 — across 7 languages, powered by Llama 3, Stable Diffusion + …
🤐Abhiroop Chintalapudi
(he/him)
GenAI Intern | M.Tech AI Student | Generative & Multimodal ML Researcher
Professional Summary
I am currently a GenAI Intern at GlobalFoundries (Bangalore), where I build AI tools designed to accelerate and optimize the PCell developer workflow. I recently completed my Master’s in Artificial Intelligence at IIT Gandhinagar, specializing at the intersection of generative models, multimodal learning, and large language models (LLMs). My academic research focused on systematically enhancing the robustness and fairness of In-Context Learning using Multi-Armed Bandits. Notably, this work successfully recovered 85–95% of baseline accuracy under non-i.i.d. distribution shifts across diverse NLP benchmarks, alongside exploratory work in fairness constraints.
Education
M.Tech in Artificial Intelligence
Indian Institute of Technology Gandhinagar
B.Tech in Artificial Intelligence
Indian Institute of Technology Hyderabad
Interests
Generative AI & Diffusion Models
Multimodal Learning (Vision-Language)
In-Context Learning & LLMs
Semantic Segmentation & Computer Vision
Reinforcement Learning & Bandits
🔬 Research & Projects
I focus on generative models, multimodal learning, and robust in-context learning. My recent work includes a story-to-video generative pipeline, multimodal fusion for physiological pain classification, and bandit-based exemplar selection for LLMs.
Current research explores multi-armed bandit partitioning for in-context learning, achieving 85-95% recovery of baseline accuracy under non-i.i.d. conditions across diverse NLP benchmarks.
Open to collaborations in Generative AI, Multimodal ML, and LLM research.
Featured Projects
Multimodal Pain Classification (m-PainAttnNet)
Extended PainAttnNet with attention-based fusion of ECG, EMG, and EDA signals, achieving 98.8% accuracy on 5-class pain intensity classification from the BioVid dataset.
Semantic Segmentation of Aerial Drone Images
Benchmarked UNet, DeepLabV3+, SegNet, and Swin Transformer on 4000×6000 px drone imagery across 23 semantic classes; UNet achieved best accuracy at 86.4%.
🏆 Achievements
- GATE (Data Science): All India Rank 1243
- CAT 2025: 99.43 percentile
- JEE Advanced: All India Rank 616
- JEE Mains: All India Rank 1037
Recent News