§ Technology
The law of the Philippines is already public. What has been missing is a way to ask it a question and trust the answer. This is the architecture that makes a machine answer the law without being allowed to make a citation up.
Every provision of the loaded laws is parsed from its official text into a single, structured corpus — each article and section addressable on its own, with its chapter and title context preserved. The source remains the authoritative published text; this is a faithful, machine-readable layer over it.
Questions are matched to provisions by meaning, not just keywords, using legal-domain embeddings. A plain-language question (“can an arrested person stay silent?”) finds the controlling provision even when it never uses those words. Layperson phrasing is also rewritten into statutory language so nothing on point is missed.
An answer is composed only from the retrieved provisions, as discrete points — each one required to cite the exact provision it rests on. The model is instructed to say plainly when the corpus does not address a question, rather than guess.
Every citation is then independently checked against the text of the provision it claims to rest on. A separate verification pass reads each statement against its cited source and marks it verified or flagged. A citation the source does not support is surfaced as a warning — the machine cannot quietly invent authority.
§ Why the firewall matters
General-purpose AI will confidently cite cases and statutes that do not exist. For law, that is disqualifying. The firewall inverts the trust model: instead of asking you to trust the model, every claim is checked against the owned source text before you see it, and anything unsupported is flagged in plain view. Authority is verifiable, not assumed.
The same corpus that answers questions is the corpus that verifies them. That is the difference between a chatbot pointed at the law and an instrument built on it. Try a question →