Inspiration
The inspiration for Atlas came from my father’s small manufacturing setup.
Growing up around it, I saw how much effort goes into running even a small factory — coordinating materials, managing production steps, handling delays, and dealing with unexpected breakdowns. What stood out to me was that most decisions during disruptions were made quickly, often under pressure, and based more on experience than structured systems.
I’ve always wanted to help improve that process, but I don’t have hands-on experience with factory operations itself. What I do have is a strong background in computing and building systems. That made me think about how software could support decision-making in environments like this.
Atlas was my attempt to bridge that gap — to take something I’ve observed closely, but don’t directly operate in, and translate it into a system that helps operators make faster, clearer decisions when things go wrong.
The goal wasn’t to replace human experience, but to support it with structured, real-time intelligence.
About the Project
Atlas is an AI decision layer designed for small manufacturing businesses. It helps operators respond to production disruptions in real time by transforming raw operational inputs into clear, executable recovery plans.
The idea behind Atlas is simple: small factories don’t fail because they lack data — they fail because they lack fast decisions when something breaks.
How I Built It
Atlas was designed as a full-stack web application with an AI-driven workflow engine at its core.
The system consists of four main parts:
A production pipeline generator, which converts natural language business descriptions into structured workflow steps. A real-time incident detection system, where disruptions can be triggered manually or through chat input. An AI decision engine, which generates three structured recovery options: Internal Fix Plan Resource Reallocation Plan External Vendor Recovery Plan A financial impact tracker, which calculates downtime cost using: Cost of Downtime=(Labor Cost+Opportunity Cost+Utility Waste)×Time Elapsed
The frontend is designed as a control-panel style dashboard with a three-column layout to reflect production state, AI reasoning, and action execution.
The backend uses a structured database (Supabase) to manage production pipelines, incidents, vendor data, and financial configuration.
Instead of dashboards or static analytics, Atlas focuses on what to do right now when production stops.
AI Integration
AI is the core of Atlas and is used in multiple ways:
Converting unstructured business descriptions into structured production pipelines Detecting which production step is affected from natural language incident reports Generating recovery strategies in real time Drafting external vendor communication when needed Interpreting non-standard CSV uploads and mapping them to financial fields
The key design principle was that AI should not just “analyze,” but actively propose decisions that can be executed immediately.
Challenges Faced
One of the biggest challenges was deciding the scope of the system.
Originally, Atlas included advanced features such as full ERP integration, predictive analytics, and multi-user collaboration. However, during development, it became clear that the most important value was in speed of response during disruptions, not system complexity.
Another challenge was designing reliable AI outputs for structured workflows. Ensuring that AI-generated production steps and recovery plans remained consistent, actionable, and non-generic required careful prompt design and constraint-based structuring.
Finally, balancing realism with hackathon constraints was critical. Many features had to be simulated (such as email sending and vendor execution) to keep the system demo-ready while preserving product credibility.
What I Learned
This project helped me understand:
How operational systems differ from consumer AI applications The importance of decision latency in real-world workflows How to design AI systems that produce structured, actionable outputs instead of free-form responses How much impact clear product positioning has on perceived value
Most importantly, I learned that in real-world systems, the value of AI is not just in intelligence — but in how quickly that intelligence can be turned into action.
Final Thought
Atlas is an early attempt at building an “AI operations layer” for small manufacturers — a system that doesn’t just show what is happening, but actively guides what should happen next.
When production breaks, Atlas doesn’t report it — it resolves it.
Future Steps
While Atlas is currently focused on real-time incident response for small manufacturing workflows, there are several clear directions for future development.
- Predictive Disruption Detection
Instead of reacting to failures, Atlas could analyze historical patterns and sensor or workflow data to predict where delays are likely to occur before they happen.
- Supplier Network Integration
A future version could integrate real supplier databases and procurement systems, allowing Atlas to not only suggest backup vendors but also check availability, pricing, and delivery times in real time.
- Learning from Past Incidents
Atlas could build a feedback loop where every resolved incident improves future recommendations, allowing the system to adapt to specific factories over time.
- Multi-Factory Scaling
The system could expand from single-factory setups to managing multiple production sites, enabling regional or distributed manufacturing coordination.
- Semi-Autonomous Execution Layer
In advanced stages, Atlas could move beyond recommendations and begin automating certain actions directly, such as reassigning resources, triggering procurement requests, or notifying suppliers automatically.
Overall, the long-term vision is to evolve Atlas from a reactive decision tool into a proactive operational intelligence system that helps small manufacturers run with the efficiency of large-scale industrial systems.
Built With
- css
- csvparsing
- javascript
- llm
- medo
- next.js
- react
- supabase
- typescript
- vercel
Log in or sign up for Devpost to join the conversation.