Inspiration
Pythia was born from a real operational friction in Arca Continental’s distribution chain: stockouts and uncommunicated product substitutions in small retail stores. Shops like “Doña María’s” depend on order accuracy to operate, yet the current system often forces blind operational decisions without customer visibility. This leads to frustration, loss of trust, and missed sales opportunities. The inspiration was to turn this breakdown point into an opportunity for transparency, trust, and long-term customer loyalty.
What it does
Pythia is a predictive intelligence system that optimizes inventory decisions at distribution centers (CEDIS) and manages product substitutions before delivery.
- Predicts the best substitution options based on historical acceptance
- Notifies customers before the truck leaves the warehouse
- Allows customers to accept or reject substitutions in real time
- Reduces logistics costs caused by rejected deliveries
- Improves overall customer experience and operational efficiency
How we built it
Pythia is designed as a dual-layer system:
- Predictive backend: analyzes historical substitution data, product rotation, and customer behavior patterns
- Propensity model: calculates acceptance probability per store and product category
- CEDIS dashboard: operational interface that suggests optimal loading decisions
- Customer mobile interface: real-time notifications and approval flow for substitutions
- Pythia Bot: conversational assistant that facilitates communication and decision-making
Challenges we ran into
- Lack of reliable data in rural routes and low-frequency products
- Balancing automation with human control in critical decisions
- Designing a system useful for both logistics teams and end customers
- Avoiding predictive recommendations feeling like forced decisions
- Translating complex predictive models into a simple, intuitive experience
What we learned
- Accuracy alone is not enough without transparency
- Involving customers in decisions increases trust and satisfaction
- Predictive models must adapt to data quality and availability
- Empathy is as important as algorithms in operational systems
- The best systems optimize relationships, not just processes
What's next for Pythia
- Expanding predictive coverage to more product categories and regions
- Improving rural prediction accuracy through better data enrichment
- Adding real-time behavioral learning from customer feedback
- Automating smarter substitution suggestions using live signals
- Scaling Pythia into a core decision-making platform across Arca Continental
Built With
- express.js
- gemini
- mongodb
- node.js
- react
- tailwind
- typescript
- vite
Log in or sign up for Devpost to join the conversation.