Market Intelligence Platform
The first production AI and data systems at a listed royalty company
OR Royalties Inc. (TSX/NYSE: OR)
The problem
A royalty company's Corporate Development team reviewed mining filings and news by hand. Material signals — the kind that move eight-figure decisions — depended on the right person happening to read the right document on the right morning. There was no platform, no data layer, and no engineering team to build one.
What I built
- Joined as the first in-house developer and owned the whole thing end to end: scoping, build, deployment, and the conversations with Corporate Development and Finance that decided what was worth building at all.
- Built the ingestion layer in Python across six live sources, pulling mining filings and news daily, with a Next.js front end deployed on Azure Container Apps.
- Ranked incoming documents by materiality rather than recency, so the review queue led with what mattered instead of what landed last.
- Built an LLM pipeline producing first-pass input for the quarterly IAS 36 impairment process, reviewed by Finance before use — a recurring process with real audit exposure, where the output is a starting point for a person rather than an answer.
- Replaced a daily manual PDF-to-print review with a zero-touch overnight pipeline, removing the step from the team's workflow entirely.
- Built a versioned database that extracts royalty, stream and offtake terms from NI 43-101 technical reports, with a human validation state on every extracted record and analysis built on top. This is the part I am still building.
- Evaluated commercial market-data vendors independently to inform the firm's data-sourcing strategy, which meant costing out the build-versus-buy case rather than just integrating what was handed to me.
Outcome
- The platform's ranked feed surfaced a material royalty buyback and a related M&A signal that the existing manual review had missed. Confirmed by Corporate Development.
- In daily use by Corporate Development and Finance.
- Retained part-time through the academic year after the internship ended, to keep building the NI 43-101 extraction database.
Stack
Note
The source is company-owned, so there is nothing to link here. The architecture notes above are the level of detail I can share; I'm happy to go deeper on the design decisions in conversation.
Next
Mosaic
Led the application team on a semester-long, eight-person build: a desktop client that had to integrate against two other teams' servers before the demo.