Dehinnet Kemi AI

• 

Inspiration

In laboratories and industrial environments, chemical risk analysis is often manual, slow, and error-prone. Accidents frequently happen due to missing historical context or a lack of proper hazard prediction systems. We were inspired to build an autonomous digital chemical safety expert that never forgets—bridging the gap between reactive safety protocols and proactive, intelligent risk management.

What it does

Dehinnet Kemi AI is an autonomous multi-agent chemical safety intelligence system. Unlike traditional AI chatbots, it operates as a coordinated team of specialized AI agents working together:

Research Agent: Conducts in-depth chemical understanding and property analysis.

Risk Agent: Handles hazard classification and severity prediction.

Environment Agent: Evaluates ecological and environmental impact.

Safety Agent: Generates PPE and safe-handling guidance.

Memory Agent: Stores and recalls past analyses.

Report Agent: Compiles final structured safety reports.

It features a persistent memory system using Supabase, allowing the AI to recall past chemical analyses and improve future decisions based on historical context. Users can trigger a full end-to-end investigation with a single click.

How we built it

Frontend: Built using Lovable AI for rapid, responsive interface development.

Database & Memory: Powered by Supabase for persistent, relational memory storage.

AI & Backend: Powered by advanced cloud LLM APIs with custom JavaScript/TypeScript backend logic.

Architecture: Designed with a memory-first multi-agent simulation architecture, where every API request cross-references historical data before generating new insights.

Challenges we ran into

Multi-Agent Coordination: Designing a multi-agent system while keeping API latency low and costs optimized.

Persistent Memory Integration: Implementing historical record retrieval without impacting real-time performance.

JSON Structure Reliability: Ensuring LLM outputs structured consistently for automated frontend parsing.

UX Balance: Simplifying a complex multi-agent reasoning flow into a seamless, intuitive one-click user experience.

Accomplishments that we're proud of

Fully Autonomous Multi-Agent System: Successfully transformed Dehinnet Kemi AI into a working, autonomous multi-agent chemical safety system that behaves like a coordinated AI safety laboratory rather than a basic chatbot.

Persistent Historical Memory: Integrated a persistent memory architecture using Supabase, enabling the system to recall past chemical analyses and continuously refine future decisions based on historical data.

One-Click Autonomous Investigation: Built a streamlined end-to-end investigation mode where users can trigger chemical safety analysis with a single click, automating memory retrieval, multi-agent reasoning, and report generation.

Production-Grade Integration: Successfully combined high-performance frontend tools (Lovable AI), cloud databases (Supabase), and state-of-the-art LLM APIs into a unified, deployment-ready platform with active user access.

What we learned

We mastered designing autonomous AI agent systems, implementing memory-backed data architectures, optimizing LLM token workflows, and transforming raw AI concepts into production-ready software solutions with real user database integration.

What's next for Dehinnet Kemi AI

Semantic Search & Deep History: Upgrading the memory system with semantic search to detect long-term safety patterns across multiple laboratory analyses.

Specialized Compliance Agents: Introducing regulatory compliance agents tailored for different regional standards and emergency simulation modules.

External Integrations: Connecting real-time chemical databases and external safety repositories to expand coverage.

Institutional Deployment: Scaling the platform for educational institutions and industrial laboratories to reduce chemical risks in real-world environments.

Built With

• agentic-ai
• ai
• artificial-intelligence
• chemical-safety
• cloud-ai
• hazard-prediction
• javascript
• llm
• lovable
• machine-learning
• multi-agent
• persistent-memory
• postgresql
• react
• rest-api
• safety-system
• supabase
• typescript
• vercel
• vite