Projects
Satellite Image Mosaicing Software
Won the Smart India Hackathon Grand Finale among 15,000 students, competing under the Indian Space Research Organization problem statement. Developed seamless satellite image mosaicing software, acknowledged by ISRO as a viable production solution — IP shared equally with the organisation.
Router-Suggest: Dynamic Routing for Multimodal Auto-Completion
Defined the Multimodal Auto-Completion (MAC) vision-language task from scratch — covering task definition, dataset curation, and evaluation protocol. Benchmarked state-of-the-art VLMs (Qwen2-VL, MiniCPM-V, PaliGemma) and proposed Router-Suggest, a dynamic routing framework achieving 2.3×–10× speedup over the best VLM baseline while maintaining near-parity accuracy.
Chat-Ghosting: Auto-Completion in Dialog Systems
Introduced the first comprehensive benchmark for chat-ghosting (predictive text completion in conversations) across 4 public dialog datasets. Proposed an entropy-based dynamic early stopping mechanism that significantly improves Partial-Precision and TES metrics, making auto-completion practical for real dialog applications.
Zero-Shot Cross-Lingual QA via Prefix Tuning on mT5
Investigated parameter-efficient fine-tuning using prefix tuning on mT5 for zero-shot cross-lingual question answering. Demonstrated competitive transfer performance with less than 1% trainable parameters compared to full fine-tuning, validating PEFT as a practical approach for multilingual NLP.
Efficient Product Embeddings & Semantic Search Engine
Showed BERT contextual embeddings outperform Word2Vec and GloVe by 10× on product retrieval across 14,000 Amazon products. Built a working semantic search engine with 3-D TensorFlow Embedding Projector visualisation for interpretability.
Publications
EACL 2026 — Industry Track
Router-Suggest: Dynamic Routing for Multimodal Auto-Completion in Visually-Grounded Dialogs
Defined the MAC vision-language task; benchmarked SOTA VLMs; proposed Router-Suggest achieving 2.3×–10× speedup with near-parity accuracy.
PDF →EACL 2026
Chat-Ghosting: Methods for Auto-Completion in Dialog Systems
First comprehensive benchmark for chat-ghosting across 4 public dialog datasets; entropy-based dynamic early stopping significantly improves Partial-Precision and TES.
PDF →On GitHub