Vistra — AI Foresight Engine
Trade-trend prediction for Dubai Customs and Zayed University.
- role
- Lead Developer
- duration
- 8 months
- team
- 3
- stack
- LLM · Python · React · Framer Motion · Exa · jsPDF
“what used to take weeks now happens in minutes.”
The problem
Government trade analysts need to synthesize signals from hundreds of disparate sources into written assessments. That process was taking weeks per brief and limiting how many scenarios could actually be tracked.
The approach
A React front-end that frames the analyst’s workflow: ask a foresight question, pick a scope, watch the engine retrieve live web sources via Exa, reason over them through an LLM layer, and emit a structured brief. Export to a government-formatted PDF with jsPDF + jspdf-autotable in one click — ready for distribution.
The hard work was the prompt/reasoning scaffolding and the visual grammar for uncertain predictions — confidence bands, evidence chips, source linking beside every claim.
The stack
Python backend for the reasoning layer, React + Framer Motion for the front-end, Exa for web search retrieval, jsPDF/jspdf-autotable for export. Enterprise-grade auth and ISO-27001-aligned data handling.
Reflection
The most important design lesson: make the model’s work legible. Every claim in the brief links back to the sources it came from; every prediction carries its confidence. That’s what made analysts trust it.