Inspiration
Interprovincial trade barriers cost Canada an estimated $14 billion a year by conservative counts, some estimates go as high as $200 billion in foregone GDP. A business that's fully licensed in Ontario can hit a wall of unfamiliar rules the moment it tries to sell in Alberta, alcohol distribution rules, meat inspection requirements, provincial professional licensing, each province runs its own rulebook and none of them talk to each other.
What made this feel urgent rather than abstract: the federal government passed the One Canadian Economy Act in June 2025, a real law creating a framework for provinces to mutually recognize each other's standards. Nobody's built the interface for it yet. We wanted to be first.
What it does
A business describes what it sells and where it wants to expand. Interbridge classifies the business into a sector, cross-references it against a database of provincial regulations, and returns a plain-language breakdown of exactly which rules apply, whether each one is a hard blocker or just a condition to satisfy, and a generated checklist to resolve it. Every claim links back to its actual source regulation and issuing authority, nothing is presented without a citation. Ambiguous or uncertain matches are flagged as low-confidence rather than resolved silently, since a wrong "you're compliant" is worse than an honest "we're not sure, check with the authority directly."
How we built it
- Backend with FastAPI and SQLite
- Two-stage LLM pipeline: one call classifies the business into a sector, a second, separate call generates the grounded explanation and checklist strictly from the regulation rows we retrieved via plain SQL, not from the model's memory.
Keeping retrieval deterministic and the generation step grounded to only the retrieved rows was the core design decision, it's what makes the citations trustworthy.
Frontend is Next.js and Tailwind, Fraunces typeface,
Built to feel like a serious civic tool rather than a consumer app.
ElevenLabs powers an optional voice intake path, since a lot of the small business owners this is built for aren't going to sit and fill out a taxonomy form correctly.
Challenges I ran into
Grounding the LLM was harder than expected, an early version would occasionally reference regulations that sounded plausible but weren't in our actual dataset. I had the backend cross-check every regulation ID the model returns against my verified database before it ever reaches the response, anything not in that verified set gets silently dropped rather than shown to the user.
I also had to change the LLM provider (from anthropic to groq) mid-build after running into API cost constraints, which meant rewriting the classification and generation calls to a different provider's API shape under time pressure without breaking the grounding logic.
Accomplishments that I'm proud of
A working end-to-end pipeline where every single regulatory claim on screen traces back to a real, cited government source, not a summary, not a paraphrase, an actual link. We also caught a live, current regulatory development while building this, the CFIA proposed amendments to interprovincial meat trade rules in July 2026, and folded that into our dataset, which meant our demo data reflected something that was still actively unfolding as we built it.
What I learned
How fragmented Canadian interprovincial regulation is, even within one sector like alcohol, four provinces means four different licensing bodies, four different label requirements, and no shared vocabulary. I a lot about LLM grounding, the difference between a model that sounds confident and a system that's actually verifiable comes down to architecture decisions like separating retrieval from generation.
What's next for interbridge
Scaling past four provinces to full national coverage, and moving from curated seed data to a maintained pipeline against real regulatory sources, ideally partnering with the Internal Trade Secretariat or provincial data. Selling into chambers of commerce, trade consultancies, and provincial economic development offices who already do this analysis by hand for their clients. Longer term, as the One Canadian Economy Act's mutual recognition framework gets implemented province by province.
Log in or sign up for Devpost to join the conversation.