ES
← Back to Portfolio
Mining & Optimization June 2026 Proprietary

Veta — Agentic Mineral-Tracking Decision Support

A private decision-support system that turns a natural-language question (typed or spoken) about a mineral-processing operation into a traceable, evidence-backed answer — a multi-stage agent pipeline routes the query to tiered solvers and grounds every recommendation in plant data with RAG.

Interface
Conversational — text + voice
Pipeline
Multi-stage agents → tiered solvers + RAG evidence
Reproducibility
Per-site, sha256-manifested artifacts
Deployment
Live (private); dual-target VPS + Azure Container Apps
Veta — Agentic Mineral-Tracking Decision Support — Architecture
#mining #decision-support #agents #rag #voice #optimization #private

Business Context

Plant decisions are high-stakes and time-sensitive, yet the supporting analysis is often locked in spreadsheets and specialists' heads. A conversational, evidence-first assistant compresses the loop from question to defensible recommendation, and makes the reasoning auditable rather than a black box.

Strategic Value

Veta is a working pattern for agentic decision support in heavy industry: a voice/text front door, a staged agent pipeline, solver tiers that spend compute only when justified, and RAG that keeps every answer grounded and traceable. The same architecture generalizes to any operation where decisions must be fast, defensible, and reproducible.

The Challenge

In a mineral-processing operation, the knowledge needed to answer "what should we do about this feed and these operating constraints?" is spread across telemetry, historical runs, and expert intuition. Pulling it together is slow, manual, and hard to audit — and an answer with no traceable evidence is hard to trust on the plant floor.

Our Approach

Veta exposes a single chat surface (text and voice). Each request flows through a multi-stage agent pipeline that interprets intent, assembles the relevant plant state, and routes to a tiered solver (cheap heuristics first, heavier optimization only when warranted). Every recommendation is grounded with retrieval over the tracked sites' data (RAG), so the answer arrives with its supporting evidence attached. An offline pipeline builds per-site artifacts that are content-addressed (sha256-manifested) for reproducibility.

Key Performance Indicators

KPIBaselineResultImpact
Question → recommendationManual analysis across tools/peopleSingle conversational surface (text + voice)Compressed decision loop
Trust in the answerOpaque, untraceableEvery answer carries its RAG evidenceAuditable recommendations

Proprietary — source code not publicly available

Architecture

veta architecture

veta architecture

A conversational front door to plant decisions

Veta is a private decision-support system for mineral-processing operations. Ask it a question — typed or spoken — and a multi-stage agent pipeline interprets the intent, gathers the relevant plant state, routes to the right solver tier, and answers with the supporting evidence attached.

This is proprietary work; the live deployment is private. The card describes the architecture and intent without exposing internal data or logic.

Technology Stack

Python FastAPI LLM agents RAG Speech-to-Text Optimization Docker Azure Container Apps

Visual assets for this project are not publicly available.