Building intelligent systems that think, see, speak, and act.

A four-agent swarm built on Google's Agent Development Kit. It ingests drone footage, Sentinel-1 SAR, and live sensor data, then thinks, speaks, and acts — warning commanders when evacuation routes turn unsafe in cities like Jakarta.

A multi-agent system that turns long YouTube videos into vertical shorts from a single URL. A 5-stage pipeline handles analysis, verification, planning, FFmpeg execution, and post-production.

Closes the interpretation gap between prenatal visits — turning smartwatch vitals, wound photos, and voice check-ins into structured, clinician-aligned risk assessments in real time.

A six-stage lifecycle from resume and JD analysis through a live bidirectional voice mock interview to a post-interview report with body language cues from webcam snapshots.

Proves transparent, step-by-step reasoning is achievable in a 2B open-source model through a disciplined three-stage training curriculum — correctness and interpretability, not parameter count.

A 6-wheeled autonomous robot that sees, understands, thinks, and speaks with emotion as a first-class input. Face → custom TFLite FER model → Gemini Robotics → Azure TTS.