Agentic AI Inventory System
Budget: $10 – $30 USD
I need a production-ready Inventory Management System that relies on autonomous, agent-driven AI to monitor, analyse, decide, and act on our inventory without constant human input while still keeping the management team fully in the loop. The core environment is Windows, so every component—from local services to any optional desktop client—must install and run smoothly there.
Key behaviours I expect
• Continuous reading of stock-level data from our existing databases or APIs, learning patterns and triggering automated actions such as re-orders or redistribution before shortages or overstock situations arise.
• A real-time dashboard that surfaces these decisions and their rationales: executive-level KPIs, drill-downs, and clear notifications whenever thresholds are met or agent actions are executed. The dashboard will be the primary way management is alerted (email or SMS could be nice-to-have later, but are not required for the first release).
• Optimisation logic that improves procurement decisions over time—think reinforcement or supervised learning that refines reorder points, economic order quantities, and supplier choice using only stock-level data at first.
• Auditability: a full log of every autonomous action, plus the ability for a human to override, approve, or roll back an agent’s decision at any stage.
• Scalability: we’re already 40+ employees across several projects; the architecture should handle more users, warehouses, and SKUs without a full rebuild.
Deliverables
1. A Windows-ready application (web, desktop, or hybrid) with autonomous inventory agents running in the background.
2. Source code, build/deploy scripts, and technical documentation detailed enough for an internal developer to maintain.
3. A hand-off session showing the system live: stock ingestion, decision simulation, dashboard alerts, and override flow.
4. A brief roadmap outlining how supplier performance and sales-trend data can be integrated later.
Acceptance criteria
– System installs on a fresh Windows machine and connects to our test dataset within one hour.
– Dashboard updates in under 30 seconds after a stock change.
– Autonomous reorder proposal accuracy meets or beats our current manual targets over a two-week pilot.
I’m comfortable with modern stacks—Python/FastAPI, .NET, Node, React, PostgreSQL, TensorFlow, LangChain, or similar—so propose what you work best with, provided it aligns with the above goals.
Key behaviours I expect
• Continuous reading of stock-level data from our existing databases or APIs, learning patterns and triggering automated actions such as re-orders or redistribution before shortages or overstock situations arise.
• A real-time dashboard that surfaces these decisions and their rationales: executive-level KPIs, drill-downs, and clear notifications whenever thresholds are met or agent actions are executed. The dashboard will be the primary way management is alerted (email or SMS could be nice-to-have later, but are not required for the first release).
• Optimisation logic that improves procurement decisions over time—think reinforcement or supervised learning that refines reorder points, economic order quantities, and supplier choice using only stock-level data at first.
• Auditability: a full log of every autonomous action, plus the ability for a human to override, approve, or roll back an agent’s decision at any stage.
• Scalability: we’re already 40+ employees across several projects; the architecture should handle more users, warehouses, and SKUs without a full rebuild.
Deliverables
1. A Windows-ready application (web, desktop, or hybrid) with autonomous inventory agents running in the background.
2. Source code, build/deploy scripts, and technical documentation detailed enough for an internal developer to maintain.
3. A hand-off session showing the system live: stock ingestion, decision simulation, dashboard alerts, and override flow.
4. A brief roadmap outlining how supplier performance and sales-trend data can be integrated later.
Acceptance criteria
– System installs on a fresh Windows machine and connects to our test dataset within one hour.
– Dashboard updates in under 30 seconds after a stock change.
– Autonomous reorder proposal accuracy meets or beats our current manual targets over a two-week pilot.
I’m comfortable with modern stacks—Python/FastAPI, .NET, Node, React, PostgreSQL, TensorFlow, LangChain, or similar—so propose what you work best with, provided it aligns with the above goals.
Related categories:
Python
Machine Learning (ML)
Node.js
PostgreSQL
API Development
FastAPI
LangChain