POV AI Indonesia — Economic Sentiment Intelligence Platform
Autonomous economic intelligence platform monitoring Indonesia's most politically sensitive indicators (BBM prices, Rupiah exchange rates) through multi-source scheduled crawling and LLM-powered analysis. Replaces hours of manual analyst research with automated, structured intelligence delivered in real time.
// Key Challenges
- Information Overload: Analysts monitoring BBM prices and Rupiah movements had to manually check dozens of news sources, government portals, and social channels — a 3-4 hour daily task.
- Source Volatility: Web scraping targets frequently change HTML structure, requiring resilient scraper architectures that don't break on minor site updates.
- Sentiment Subjectivity: Raw news headlines don't capture real public sentiment — LLM analysis needed calibration to distinguish factual reporting from emotional public reaction.
- Data Freshness: Economic conditions change rapidly; a 24-hour analysis lag makes intelligence irrelevant for decision-making during fast-moving events.
// Approach & Solutions
1. Multi-Source Adaptive Web Crawler
CSS selector-based scraper with fallback XPath patterns targets 15+ financial and government sources. Scraper health monitoring detects source structural changes and alerts for re-calibration, preventing silent data gaps.
2. LLM-Powered Structured Analysis Pipeline
Scraped raw content passes through a structured LLM prompt pipeline generating: (1) factual price/rate summary, (2) public sentiment score (-5 to +5), (3) trend direction assessment, and (4) executive conclusion — all in under 8 seconds per cycle.
3. Cron-Scheduled Intelligence Cycles
Configurable cron scheduler runs crawl-analyze-store cycles at user-defined intervals (hourly to daily), ensuring intelligence freshness is tuned to the volatility of each indicator being monitored.
// Tech Stack
| Technology | Reason |
|---|---|
| Python + BeautifulSoup / Requests | Python's rich scraping ecosystem enables rapid adaptation to new target sources; CSS selector + XPath dual-strategy maintains resilience against minor HTML structure changes. |
| LLM via OpenAI / OpenRouter | Structured output prompting (JSON mode) ensures consistent machine-parseable analysis results across thousands of crawl cycles without manual quality control. |
| FastAPI + PostgreSQL | FastAPI serves real-time intelligence API endpoints to the frontend dashboard; PostgreSQL stores historical trend data enabling multi-week sentiment trajectory analysis. |
| Docker + Cron | Containerized scheduler ensures consistent execution environment across deployments; cron-based scheduling provides predictable intelligence refresh cycles without infrastructure overhead. |
// Results & Business Impact
- 15+ - Data Sources Monitored (Financial media, government portals, and public discussion channels)
- <8s - Analysis Cycle Time (From raw crawl to structured intelligence summary per cycle)
- 100% - Autonomous Operation (Zero manual analyst intervention required for routine intelligence cycles)
- Live - Production Status (Actively monitoring Indonesian economic indicators at idn.povai.my.id)
// Project Gallery
