AN MINH
← Back to Projects

AI / Data Pipeline

Multi-source Text-based Coffee Price Forecasting System

A forecasting pipeline combining historical coffee prices with signals extracted from 30,000 Vietnamese news articles for next-day price prediction.

Client
Team Project (Size: 5)
Role
AI & Data Engineer
Stack
Python, HuggingFace, PhoBERT, Pandas, NumPy, FastAPI, React
Date
Feb 2026 — May 2026
Repository
Coffee Price Forecasting dashboard
Fig 1.0 — Coffee Price Forecasting dashboard

Technological & Algorithmic Highlights

Multi-source Signal Extraction

Aggregated and processed historical coffee prices along with sentiment and signals extracted from over 30,000 Vietnamese news articles using NLP.

LLM Workflow & Evaluation

Developed an LLM workflow for next-day price prediction, conducting comparative analysis between DeepSeek and Gemini outputs.

Operation Flow

  1. 01Daily coffee prices and Vietnamese news articles are ingested.
  2. 02News articles are processed through PhoBERT to extract market sentiment and signals.
  3. 03Historical prices and text signals are fused into the forecasting pipeline.
  4. 04The LLM workflow predicts the next-day coffee price.
  5. 05Results and visualizations are presented via the FastAPI/React dashboard.

System Infrastructure

A decoupled architecture where a heavy Python-based AI pipeline processes data and serves predictions via FastAPI to a React frontend.

Infrastructure parameters

  • • Backend API: FastAPI
  • • Frontend: React
  • • Models: HuggingFace, PhoBERT
  • • Data Stack: Pandas, NumPy