My Craziest Side Project Yet: Drawva

Meet Taqui, a fullstack web developer who recently delved into the world of Agentic AI and GenAI to create an ambitious project called Drawva. Drawva is an innovative infinite canvas whiteboard engine that leverages multimodal AI perception to understand hand-drawn gestures and transform them into interactive diagrams, code widgets, and math equations.

Rather than simply rendering static pixels, Drawva's core lies in its DeepSeek Harness Agent Runtime, built atop a Cordis microkernel runtime. This system enables multi-turn tool calling through typed contracts, token metering, and effective compaction to maintain agent context while minimizing costs during prolonged sessions.

To bridge the gap between freehand 2D canvases and LLMs, Taqui employed custom prompt engineering techniques. One such method, the Coordinate Contract, ensures precise mathematical pixel-to-global transformations from visual snapshots. Meanwhile, the Zero-Heading Redundancy approach prevents the AI from outputting duplicate title banners, preserving a natural co-authored canvas appearance.

Drawva also boasts a unique Hybrid Wireframe Utilization feature. When users sketch a 3-box wireframe, the AI leaves their pen lines intact while rendering live data directly within the drawn boxes, all with transparent backdrops. Moreover, the Two-Command In-Place Replacement allows for seamless transitions from rough sketches to interactive applets by first emitting an atomic erase command, followed swiftly by the mounted widget.

The platform offers native sketchnoting capabilities through a dedicated mode using pure vector strokes to create 3D loss surfaces, neural net graphs, and complex planes. Drawva also supports extensible on-demand plugins, with 11 currently available, ranging from Live Data (weather, stocks, tech news, etc.) to Developer & Media integrations (GitHub Pulse, Image Search, Flowcharts), and Diagram Engines (Mermaid, Graphviz DOT, Vega-Lite charts, SMILES molecular structures, BPMN, Cytoscape, GeoJSON) accompanied by MathJax LaTeX typesetting.

The entire Drawva project is open-source and hosted on GitHub, where Taqui tracked features, plugin contracts, and releases. The project was built using Next.js 16, React 19, Vercel AI SDK, Tailwind CSS v4, and Base UI, with everything autosaved to IndexedDB for seamless collaboration via WebRTC in peer-to-peer mode.

Taqui's journey to develop Drawva came at a pivotal time, as the tech market faced uncertainty, and traditional coding skills seemed threatened by AI advancements. However, this experience proved that developers could evolve by embracing Agentic AI and building ambitious tools to address real-world problems. Taqui invites everyone to explore Drawva, experiment with their sketches, and share their creations on the community page.

With any luck, this open-source tool might inspire a spark in other developers, reigniting their passion for building.