Cloudglue and Tinycloud
Cloudglue APIs and Tinycloud agents for turning video collections into searchable, queryable knowledge with replayable evidence.
Co-founder and CTO
Hi, I’m Kevin Dela Rosa.
I make video useful.
For people and AI.
I build systems that search, understand, and work with video. Co-founder and CTO at Cloudglue. Previously, I led ML infrastructure and shipped computer vision at Snap.
Research, products, and the occasional computer-vision rabbit hole.
Selected work
What I’m building now, from production systems to personal experiments.
Overcast
A video OSINT agent that searches, listens, matches, maps, and remembers across a collection of video.
Creator
mDex
An offline card scanner and open-source visual matcher. A computer-vision rabbit hole, built for a real card show.
CreatorEarlier work: Snapchat Scan & visual searchML infrastructure at Snap
From the notebook
The ideas, experiments, and decisions behind the work.
I put every CSI button in a CLI
Building Overcast: the real algorithms behind the hacker-movie tropes, connected to an agent that remembers.
I vibe coded an offline card scanner for the National
10,023 trading cards, an offline iPhone scanner, and a return to classic computer vision.
Research, shared
From video retrieval to the systems that put it to work.
Tool-Skills vs. Workflow-Skills: Routing and Composition in a Multimodal Agent Harness
Case study of Tinycloud's multimodal video agent harness, showing how tool-skills and workflow-skills affect routing and composition.
Smart Routing for Multimodal Video Retrieval: When to Search What
Routes queries across ASR, OCR, and visual indices to reduce retrieval cost while preserving effectiveness.
Prefer a walkthrough? Watch conference talks and demos.
A builder at heart.
I’ve worked across research, consumer products, and the infrastructure that connects them. These days I’m building at Cloudglue and exploring what agents can do with video. Sometimes that leads to a conference demo; sometimes, an offline scanner for my trading cards.