Turning Tasks into dynamic Systems

Publicly-maintained intelligent document analyzer. Extracts text (OCR), summarizes, and rewrites complex documents into concise, readable content. Built with modern NLP pipelines for instant extraction and summarization.

Customer Acquisition Attribution Analysis. ImpactIQ analyzes the Criteo Uplift dataset to predict customer conversion probabilities and explain feature attributions using SHAP (SHapley Additive exPlanations) values.

Agentic multi-modal role-based meta prompting architecture. Built to assist in the creation of AI prompts for various tasks for professionals.

Semantic Segmentation of Aerial Imagery. Open-source toolkit for pixel-wise land cover classification on satellite images. Employs a U-Net architecture for precise segmentation and robust generalization across diverse geographies.

High-performance n-gram language models optimized for low-resource hardware. Implements probabilistic sampling, smoothing, and parallelized processing for efficient on-device text generation.

Empirical study of low‑data learning across text, image, and audio. Proposes modality alignment and adaptive fine‑tuning strategies that improve generalization under resource constraints.

• Ablations on meta‑learning adapters (few‑shot gains)
• Cross‑modal regularization (zero‑shot robustness)
• Dynamic prompting for unseen classes