WeeFarm — AI-Assisted Decision-Support Platform

Overview

Project context

Built an evidence-first assistant workflow combining SQL factual analytics, RAG contextual retrieval, ML advisory signals, and structured response composition with human manager review.

Problem

What needed to be solved

Cooperative operations required structured decision support across stock, lots, traceability, treasury, and exports while preserving factual evidence and manager oversight.

Impact

Delivery outcomes

• Built with FastAPI, Next.js, PostgreSQL/Supabase, pgvector, SQLAlchemy, Alembic, Docker, Azure Container Apps, and Vercel.
• Validated baseline behavior with 20/20 executed cases, 17 PASS / 3 PARTIAL / 0 FAIL, 100% route accuracy, and 0 runtime errors.
• Integrated readiness-gated ML advisory signals with strict train/test validation, reaching 0.8412 high-risk recall.

Architecture

Execution flow

Step-by-step implementation path used to deliver this project.

1. Step 1

   Operational PostgreSQL/Supabase data

2. Step 2

   SQL factual layer

3. Step 3

   RAG contextual layer

4. Step 4

   ML advisory layer

5. Step 5

   LLM response composition

6. Step 6

   Human manager review

Features

Capabilities shipped

Validation

Measured evidence

• 20/20 executed assistant audit cases
• 17 PASS / 3 PARTIAL / 0 FAIL
• 100% route accuracy
• 0 runtime errors
• 0.8412 high-risk recall

Media

Screenshots and recordings

WeeFarm — AI-Assisted Decision-Support Platform screenshot 1

Limitations

Scope boundaries

• Validation-backed prototype; not positioned as production-ready enterprise automation.
• ML signals are advisory and readiness-gated rather than autonomous control.
• Human manager review remains required before operational execution.

More projects