Coding agents now write more code than anyone reads. We work
spec-first: define the behavior, derive the tests from the
spec, and ship only when it matches. Rooted in the Chalmers
formal-methods tradition, and proven on our own products, Ripenn and SqlProof.
// methods
[spec-first]
We write the specification before the code. Behavioral invariants define what the system must do; property-based tests enforce them across inputs nobody thought to try, and mutation testing checks that the suite would notice if the code were wrong. AI outputs are verified against the spec, not eyeballed.
[agent-specs]
Agents write code faster than anyone can read it. We build the skills, harnesses, and review gates that give coding agents a specification to follow and a test oracle to answer to — so their output is reproducible, auditable, and correctable when it diverges.
[product-development]
End-to-end product work grounded in the same method. Ripenn, our AI search visibility platform, is built on it: every score is backed by the stored answer that produced it, so a brand can see not just where it stands in AI answers but why.
spec-first development, behavioral invariants, and specification coverage —
for AI systems where correctness is a requirement.
// artifacts
A lab is defined by what it publishes. We run our own products on the same method we bring to engagements.
Property-based testing for PostgreSQL and Supabase, built on Hypothesis. Generates hundreds of valid datasets from your schema, runs your queries and RLS policies as different users, and shrinks any failure to the smallest reproducer. Catches the SQL bugs your coding agent wrote.
SqlProof Cloud — hosted scenario databases and pull-request review on top of it — is in private pilot.
sqlproof.com →A Claude Code plugin that makes coding agents write real property-based tests. Most agent-written properties are tautologies or example tests in disguise; the skill forces the property to be stated in English and the oracle named before any generator is written. Ships with a public, reproducible eval.
github.com/alialavia/pbt-skills →AI search visibility platform. Tracks how ChatGPT, Claude, Perplexity, and Gemini represent a brand, keeps the full answer behind every number, and explains which prompts, engines, and sources drive it. Measurement you can audit, not a score you have to trust.
ripenn.ai →
// founder
25+ years shipping software in public safety (Axon — internal engineering awards), autonomous vehicles, and healthcare.
MSc, Secure and Dependable Computer Systems, Chalmers — the academic home of QuickCheck and property-based testing (John Hughes). PhD, Human-Computer Interaction, ETH Zurich.
Co-chairs Justice Rewired, an AI + criminal-law initiative at Osgoode Professional Development, Toronto.
LinkedIn →// engage
If your AI product or coding-agent workflow has to be right, not just plausible, we would like to hear about it.
