AI & Automation
Live demoASAN MACRO — AI-Powered Trade Intelligence Agent
Tool-augmented LLM agent that converts structured trade data into grounded, decision-ready economic sentiment reports.
Role: Sole builder (data layer, agent, evaluation, and reporting)
Visual evidence
Embedded preview of the working product. The iframe stays off until you load it. Open full-screen for the complete walkthrough.
The live demo is a third-party site. It is not loaded until you ask, so this page stays fast and private by default.
Situation
Manual Comtrade-to-spreadsheet-to-generic-LLM workflows are slow, inconsistent, and poorly grounded for economists and strategy readers who need repeatable briefs.
Decision question
Tool-augmented LLM agent that converts structured trade data into grounded, decision-ready economic sentiment reports.
Data and constraints
See method and results below for sample, coverage, and documented limits.
Method
Designed an agentic workflow over SQLite trade data with custom analyst tools and lightweight RAG. The system supports multi-backend LLMs, optional Comtrade ingest, structured report formats, evaluation harnesses, and adversarial/safety probes so claims stay tied to stored flows rather than free-form guessing.
Tools
- Python
- SQLite
- LLM tool-calling / RAG
- UN Comtrade ingest scripts
- Evaluation & adversarial test harnesses
Deliverables
- CLI trade sentiment report generator
- Nine analyst query tools over trade flows
- HTML / LinkedIn / JSON output formats
- Evaluation and safety probe scripts
Outcome / decision value
- Built an end-to-end agentic product for BRICS and sector-focused trade analysis with documented evaluation scaffolding.
- Public demo deployed at asan-macro.vercel.app for trade sentiment briefs grounded in UN Comtrade, World Bank, and IMF series.