Inspiration
What it does
How we built it# โก OTTO-TASK: SHADOW MULTIMODAL AGENT โก
๐ Inspiration
Most marketplaces are static digital shelves. We wanted to create a Cyberpunk Autonomous Bunker where the interface isn't just a tool, but an entity. Otto-task was inspired by the need for high-security, intelligent asset management in decentralized environments, where an AI agentโShadowโacts as the ultimate gatekeeper and navigator.
๐๏ธ How we built it
The project is built on a decoupled architecture to ensure maximum modularity and speed:
- Neural Core: Powered by Gemma-3-27B-IT via the Google GenAI SDK.
- Backend Infrastructure: A robust Django framework manages the "Bunker's" logic and security.
- Dynamic Grounding: We implemented a real-time data extraction layer that feeds structured SQL data (Brand, Model, Status) directly into the LLM's context window.
- Storage: Cloudinary for multimodal asset handling (Images/Tech specs).
- Environment: Deployed on Render/Google Cloud for 24/7 operational status.
๐งช Challenges we overcame
The biggest hurdle was the "Identity & Hallucination Gap." Initially, the agent would recognize a "Cell phone" but fail to identify specific technical specs like "Xiomi A15." We solved this by building a Deep Attribute Extraction protocol. Instead of simple prompting, we forced the system to perform a "System Scan" of the database attributes, mapping natural language queries to specific ID fields. We also synchronized two separate repositories (Otto-market and Otto-task) to work as a unified neural network.
๐ What we learned
We mastered the art of RAG (Retrieval-Augmented Generation) without the overhead of complex vector DBs for mid-sized inventories. We learned that Gemma-3 is exceptionally capable of maintaining a "Cyberpunk Persona" while simultaneously processing rigid technical data. We also refined our skills in Multimodal Data Mapping, ensuring that every asset in the bunker has a verifiable digital twin in the database.
๐ ๏ธ Technical Specs (The Math of Shadow)
Shadow calculates inventory health using real-time aggregation: $$V_{total} = \sum_{i=1}^{n} P_i$$ Where $V$ is the Total Bunker Value and $P$ represents the price of each active asset $i$. This ensures Shadow always knows the exact value of the operation.
OPERATOR: Dmfhdilyd | STATUS: MISSION READY ๐ข
Log in or sign up for Devpost to join the conversation.