Manuflow

AI inventory management system for small and medium-sized manufacturers

Comment

Manuflow (Startup Track)

Inspiration

Both of our families have been in the manufacturing business for over a decade. We have both seen firsthand how inventory chaos can make or break a factory. One recurring problem that fascinated us was how seasonal demand patterns and weather changes would catch manufacturers off-guard—leading to either costly overstock or lost sales from stockouts.

What it does

For this hackathon, we built a demand forecasting tool that leverages Snowflake's data marketplace to access external weather data and correlate it with historical sales patterns.

Key capabilities:

  • Voice-first interface: Warehouse workers can simply ask "Which products are weather-sensitive?" or "Forecast milk demand for the next 2 weeks based on temperature"
  • Temperature correlation analysis: Identifies which products sell more in hot vs. cold weather using statistical regression analysis
  • Forward-looking forecasts: Uses 14-day weather forecasts to predict demand changes and inventory shortages
  • Weather risk alerts: Proactively identifies products at risk of overstock or stockouts based on upcoming weather patterns
  • Regional insights: Compares temperature sensitivity across different warehouse locations

How we built it

Architecture

Data Foundation (Snowflake)

  • Migrated inventory and sales transaction data from Supabase to Snowflake using automated cron jobs
  • Integrated weather data from Snowflake's Data Marketplace (historical temperature records for San Francisco)
  • Added 14-day weather forecast data to enable future demand predictions
  • Built optimized SQL queries using CORR(), REGR_SLOPE(), and window functions for correlation analysis

Intelligence Layer (AWS Lambda)

  • Built middleware with 6 different insight types: product correlation, demand forecasting (historical & future), regional sensitivity, weather risk, and temperature thresholds
  • Implemented JWT authentication for secure Snowflake API access
  • Created voice-optimized response formatting that translates complex statistical analysis into natural language

Voice Interface (ElevenLabs)

  • Configured custom tools with structured parameters for different analysis types
  • Designed conversational prompts that guide warehouse workers through complex queries
  • Integrated real-time query execution with sub-10-second response times

API Layer (AWS API Gateway)

  • Set up RESTful endpoints with CORS support
  • Implemented dual routes: basic inventory queries and advanced temperature insights

Tech Stack

  • Data Warehouse: Snowflake (compute, storage, analytics)
  • External Data: Snowflake Data Marketplace (weather data)
  • Middleware: AWS Lambda (Node.js)
  • Voice AI: ElevenLabs conversational agent

Challenges we ran into

Initially struggled with matching SKUs between our inventory system and sales transactions, which caused join failures.

Solution: Built comprehensive data validation scripts and added missing products to ensure referential integrity.

Accomplishments that we're proud of

This has been the most demanded feature out of all our customer interviews, so we are excited to test it out with our pilot program after this hackathon.

What we learned

  • Snowflake's analytical functions (CORR, REGR_SLOPE) are incredibly powerful for statistical analysis at scale
  • JWT authentication with RSA keys is complex but provides enterprise-grade security
  • Voice interfaces require fundamentally different UX design than visual dashboards—conciseness and clarity over completeness

What's next for Manuflow

We are going to launch our pilot program after this hackathon and continue iterating the product while trying to raise our pre-seed fund.

Built With

Share this project:

Updates