Precedent you can
actually rely on.
Every sentence traces to a paragraph in a real judgment and Prima declines to answer when the archive cannot support one.
Expensive to find. Dangerous to get wrong.
Days of billable time
Finding the authority that fits the facts is manual, keyword-driven work. Junior associates spend days on it, and clients are billed for every hour.
No way to check it still stands
A judgment can be overruled, distinguished or doubted years later. In East Africa there is no citator — no reliable way to ask whether an authority is still good law.
AI has already burned the profession
Lawyers have been sanctioned for filing submissions containing citations a model invented. Trust is now the barrier, not capability.
A case that has been overruled is not a precedent. It is a liability.
Westlaw and LexisNexis solved this for the United States forty years ago. Nobody has solved it for Uganda, Kenya or Tanzania.
Thirteen years later a higher court decided the opposite. The judgment still sits in every archive and still tops a keyword search. A lawyer relying on it walks into court with a losing case and doesn't know it.
Three layers. Only one of them is search.
The archive
A normalised, citation-linked corpus of East African and Commonwealth judgments — cleaned, structured, machine-readable. The asset nobody else has built.
The citator
A treatment graph across every judgment: followed, applied, distinguished, doubted, overruled. This is what tells a lawyer whether an authority is safe to rely on.
The reasoning layer
Semantic retrieval, case analysis and adversarial testing over the top — every claim bound to a pin cite in the archive beneath it.
Competitors ship layer three. Layers one and two are the business.
What it does for a working advocate
Semantic precedent search
Describe the facts in plain language. Prima retrieves authority on meaning, not keyword overlap.
Treatment check
Every authority returned carries its history: what has followed it, doubted it, overturned it.
Adversarial review
Upload opposing counsel's submission. Get every authority they rely on, its weaknesses, and counter-authority.
Matter memory
Pleadings, contracts, discovery and correspondence ingested per case. The system reasons over your file, not just the law.
Outcome base rates
How this cause of action has historically resolved in this court, and which arguments correlated with success.
Drafting with locked citations
Submissions drafted against the archive, where no citation can exist unless it exists in the record.
Every other tool argues your side
What everyone builds
- Find me authority that supports my position
- Returns a ranked list of documents
- Leaves the lawyer to imagine the counter-argument
- Confirms what you already believed
What Prima does
- Builds the strongest case against you first
- Names the authority opposing counsel will rely on
- Flags where your file is thin before the judge does
- Returns an argument map, not a document list
A litigator does not need a search engine at 11pm. They need to know what is coming.
Refusal is the feature
Pin cite or nothing
Every sentence Prima produces resolves to a paragraph in a real judgment. No claim survives without a source the user can open.
Grounded in the archive only
The model reasons over retrieved text. It is not permitted to answer from parametric memory, which is where invented citations come from.
It says when it cannot answer
Where the corpus does not support a proposition, Prima declines rather than improvising. Silence is safer than a sanction.
The profession does not need a more confident model. It needs one that can be checked.
East African courts do not reason in isolation
Ugandan, Kenyan and Tanzanian courts routinely rely on persuasive authority from across the common law world. A tool trained on one jurisdiction cannot follow that reasoning.
Cross-jurisdictional retrieval is not a feature a US-built product can retrofit cheaply. It requires the corpus, the citation conventions and the doctrinal lineage to be modelled together — which is exactly the work this project is doing anyway.
Why this is hard to copy
The corpus is grunt work
Regional judgments arrive as inconsistent scans with no standard citation format. Normalising them is slow, unglamorous and unavoidable. Well-funded competitors will not do it.
The citator compounds
A treatment graph gets more valuable with every judgment added and every correction made. It cannot be bought, only accumulated.
Jurisdiction is the barrier
Global vendors optimise for the largest markets first. East Africa is not on that roadmap — and by the time it is, the archive is already built.
Trust transfers slowly
Once a firm has relied on a citator in front of a judge and been right, switching cost is professional, not technical.
Built on a stack chosen for verifiability
Ingestion
Judgments normalised, segmented by paragraph and assigned stable citation identifiers.
Embedding
VoyageAI embeddings over paragraph-level units, so retrieval resolves to a pin cite rather than a document.
Index
LanceDB vector store holding the archive and the treatment graph alongside it.
Reasoning
Anthropic models constrained to retrieved context, with refusal on unsupported propositions.
Delivery
FastAPI service and a deliberately simple interface — the product is the answer, not the chrome.
Retrieval is scoped to the paragraph so that every generated sentence has somewhere to point. The architecture enforces the trust claim rather than asking users to take it on faith.
Four buyers, four reasons
Litigation firms
Leverage. One associate covering the research load of three, with the citation risk removed.
In-house counsel
Cost control. Fewer hours outsourced to external firms for research that can be done internally.
Judiciary & registries
Consistency. Judges and clerks checking precedent against a maintained treatment record.
Law faculties & legal aid
Access. Students and underfunded practitioners reaching authority they currently cannot afford to find.
Firms buy leverage. Institutions buy consistency. Faculties buy access. The same archive serves all four.
Priced for a market that will not pay Harvey rates
Research
- Semantic precedent search
- Full archive access
- Pin-cited answers
- Individual advocates and small chambers
Citator
- Everything in Research
- Full treatment history
- Overruled and doubted alerts
- The tier the risk-conscious firm buys
Matter
- Everything in Citator
- Adversarial review
- Matter ingestion and drafting
- Priced against a single billed hour
Per-matter pricing beats per-seat in a region where firms are small and cases are the unit of work.
Archive first. Everything else follows.
Build the archive
Ingest and normalise Ugandan judgments end to end. Paragraph-level segmentation, stable citations, clean retrieval.
Ship the citator
Extract treatment relationships between judgments. This is the moment the product stops being search.
Open the matter file
Per-case ingestion, adversarial review and drafting. The paralegal, not the library.
Extend the region
Kenya and Tanzania, then Commonwealth persuasive authority. The same pipeline, more jurisdictions.
Nothing in phases two to four is possible without phase one. That is the whole strategy.
IUEA is an unusually good place to build this
A Faculty of Law on campus
Doctrinal review, annotation and treatment tagging need trained legal readers. The university has them, and they have every reason to care about the result.
A research mandate, not a sales quota
The archive takes years to be worth what it will be worth. A university can hold that horizon; a startup burning runway usually cannot.
Institutional standing
Judiciaries and registries share archives with chartered universities more readily than with vendors. Access is the hardest input to buy.
Engineering already in place
The AI systems capability at IUEA is operational, not theoretical. This is an extension of work already running in production.
Prima is not a product looking for an institution. It is an institution's advantage turned into a product.
The workspace view in context
Embedded from iuea-prima-product-1.html.
Answers, evidence and treatment history
Embedded from iuea-prima-product-2.html.
Digitising Uganda's law, judgment by judgment
Embedded from iuea-prima-archive-1.html.
The Ratio Engine — meaning and structure, indexed together
Embedded from iuea-prima-archive-2.html.
Back the archive
Search is the demo.
The archive is the business.