People ask why I am running two ventures at the same time, in two markets that look nothing alike. Mneme HQ is an architectural governance layer for AI-assisted development. CannabisDealsUS is a pricing and deals platform for the United States cannabis market. On the surface, those are different companies for different audiences.

I see them as the same bet placed twice. Both are wagers that in markets where the surface looks chaotic, the durable value lives in the data layer underneath. The product is whatever lets the buyer act on that data without having to assemble it themselves.

“The interesting work is not the interface. It is the schema that makes the interface possible.”

What the two markets have in common

The first thing both markets share is that the information already exists, somewhere, in some form, but nobody has done the work to make it queryable. In AI-assisted development, the decisions a team has made about its architecture exist — in Slack threads, in pull request comments, in the heads of senior engineers. They are not in a form that an AI agent can read. In US cannabis, dispensary menus exist — on websites, in point-of-sale systems, in regulatory filings — but the prices, strains, deals, and inventory are not in a form anyone can query across states.

The second thing they share is regulatory complexity that punishes lazy modelling. AI governance has compliance shadows around it: data residency, audit trails, accountability for what an automated system produces. US cannabis has fifty different bodies of state law, varying tax structures, product-form rules, advertising restrictions, and reciprocal-recognition rules between medical and adult-use programmes. In both cases, you cannot ship a useful product without a model of the regulatory surface. A spreadsheet is not enough.

The third is that the obvious move — scrape, dedupe, ship — produces a product that breaks within a quarter. The interesting move is the one that survives the second quarter.

What I learned building one that I am using in the other

CannabisDealsUS came first. The work on it taught me three things that ended up shaping Mneme.

Schema is destiny. The hardest decision in the cannabis platform was how to represent a strain. There is no canonical strain catalogue. Producers name the same genetics differently; consumers ask for them by different names again. We had to commit to a normalised representation early, knowing it would be wrong in interesting ways and that retrofitting would be expensive. Every product feature we shipped afterwards lived or died by that choice. Mneme’s decision schema came from this lesson — if you let a team write decisions in free text, the platform cannot reason about them later.

State is the product. The cannabis platform’s value is not the scrape; it is the historical pricing curve, the deal-availability time series, the regional comparisons. None of that exists without persistence. Mneme is the same thought applied to engineering decisions — the value is not the rule, it is the record of when it was made, who made it, what code it touches, and how it has evolved.

The boring infrastructure compounds. In cannabis data the unglamorous wins were the things that mattered: deduplication, address normalisation, rate-limited scrapers that survive site changes, a schema migration story that did not break the dashboard. In Mneme the equivalents are repo integration, deterministic decision injection, CI plumbing. Nobody buys a tool for boring infrastructure, but nobody keeps using a tool whose boring infrastructure breaks.

Why now for cannabis

The US cannabis market is at the point where the early-stage scramble is over and the consolidation is starting. There are roughly 23 adult-use states; the patchwork is stable enough to model and unstable enough that the modelling itself is a moat. A consumer in a legal state has dozens of dispensary options within driving distance, with prices that vary by more than they should and deals that rotate weekly. There is no national, neutral source of truth.

The opportunity is the kind I find most interesting: messy public data, real consumer demand, a market that nobody else seems to want to do the unglamorous work for. The fact that the topic raises eyebrows at dinner parties is, in my experience, a feature.

What this means for theovalmis.com

I am going to write about both ventures here. Most of the AI governance audience does not care about cannabis pricing and vice versa, but the engineering and data-systems audience cares about both — because both are case studies in the same set of problems. Schema design under regulatory uncertainty. Multi-source ETL with hostile sources. Decision infrastructure as a product. Operating a small founder team without a tech monoculture.

If you came here for Mneme, the cannabis posts will give you a different angle on the same engineering problems. If you came for cannabis data, the Mneme posts will explain how I think about the underlying systems work. The thesis is one thesis. Two products are how I am testing it.