I make machines behave.

A multimodal agent runtime that coordinates speech recognition, language generation, voice, facial animation and gesture. It connects models and interactive clients through shared interfaces to support responsive embodied conversations.

A modular toolkit for assembling interactive virtual humans from motion processing, behavior generation, agent runtimes and rendering clients. It brings independent components—including Buraq—together to make research systems easier to integrate, reuse and deploy.

Motion-data infrastructure that converts diverse datasets, skeletons and file formats into a shared representation with traceable provenance. It reduces repeated preparation work and supports consistent training, evaluation and animation across motion projects.

An open research atlas connecting co-speech gesture-generation methods with human gesture theory, datasets, evaluation metrics and application evidence. Structured comparisons and citation relationships help researchers trace ideas, compare approaches and find the resources needed to build on prior work.

A framework for generating coordinated speaking and listening behavior during continuous two-way conversation. It connects body motion with turn-taking, backchannels, interruptions and pauses to support agents that participate while another person is speaking.

A framework for converting ordinary monocular video into complete, clean and retargetable human motion. It connects visual observations with learned motion representations to make motion acquisition more accessible and economical.

A synthetic-data platform that renders skeletal motion into video alongside corresponding joint positions, camera geometry, segmentation and visibility information. It supplies controlled observations and known ground truth for learning and evaluating motion-reconstruction methods.

A review of the methods and practical costs involved in turning video and motion recordings into usable research corpora. It examines compute, labor, failures, licensing and reporting practices to clarify the resources needed for accessible motion research.

A reusable evaluation framework for comparing motion-generation methods and baselines through consistent data interfaces and measurement protocols. It brings motion quality, semantic relevance, diversity and computational cost into a common evaluation workflow.

Research into evaluation methods that capture the perceptual quality, meaning and behavioral appropriateness of co-speech gestures. It examines whether automatic measurements agree with human judgments and remain reliable across datasets and recording conditions.

A text-conditioned gesture framework with interpretable controls over movement characteristics such as handedness, height, amplitude, speed and emotional expression. It gives authors direct influence over a virtual human’s delivery while preserving the relationship between speech meaning and motion.

A research program examining how consistency between a digital persona’s identity and behavior shapes human perception. Controlled comparisons of behavioral cues help explain when a virtual character feels believable, appropriate and consistent with its claimed identity.

A shared motion representation for transferring movement between characters with different skeletal structures and proportions. It aims to preserve the meaning and character of a performance while reducing dependence on rig-specific retargeting procedures.

A gesture framework that adapts movement timing and gesture phases to synthesized speech using reference examples. It addresses changes in voice, speaking rate and delivery so that meaningful movements remain coordinated with the spoken message.

An investigation of speech-recognition representations as conditioning signals for co-speech gesture generation. It examines how linguistic content and acoustic information contribute to movement, seeking representations that preserve meaning while supporting expressive delivery.

A structured language for describing relationships among the hands, body, movement and speech, coupled with a description-to-motion framework. It makes gesture intent more explicit so that motion can be generated, inspected and controlled through understandable descriptions.

A motion-learning framework that combines gesture and action-language tasks through a shared token-based model. It investigates whether knowledge transferred between tasks improves the semantic relevance and generalization of generated motion.

A framework that derives gesture generation from representations inside speech-synthesis models. By sharing information between voice and movement generation, it investigates how embodied delivery can follow speech content and prosody more closely.

A text-and-motion retrieval framework that selects recorded gestures using speech meaning and reference examples of a speaker’s movement style. It aims to combine captured-motion realism with responsive, personalized behavior for unfamiliar speakers.

A lightweight framework that combines speech-conditioned emotional expression with subtle liveness movements around the eyes and eyebrows. It complements independent lip synchronization to make virtual agents more expressive and lifelike during conversation.