About Unknown Verdict
A retrieval-first legal engine for Indian law. Verbatim statutory text, cited to the page. No generation. No advice. No hallucination.
What it is
Unknown Verdict indexes official Indian legal documents — statutes, rules, and forms — into a full-text search engine. When you ask a question, it returns the exact text of the relevant provisions, with the source filename and the page number from the original PDF. Nothing is paraphrased. Nothing is summarised. Nothing is generated.
The engine does not use a large language model in the retrieval path. There is no completion. There is no prompt. There is only a search index and a structured match against section labels, source documents, and text tokens.
Why retrieval-only
Every legal AI product that generates prose will, eventually, produce a fabricated citation. This has already happened — courts in the United States have sanctioned lawyers for filing briefs containing cases invented by an LLM. Indian courts will follow.
The problem is structural: a generative model answers from probability, not from source. Even with retrieval-augmented generation, the model is free to reformulate, paraphrase, or extend. Any of those can become a misquote.
Retrieval-only removes that risk at the architecture level. If the text is not in the index, the engine returns nothing. It cannot invent.
Design principle: the engine is a mirror, not an author. What it shows you is what the statute says. Nothing more, nothing less.
Design principles
- Verbatim only. Every response is a direct quote from an indexed source.
- Cited to the page. Every chunk carries its source filename and page number.
- Zero hallucination. The engine has no capacity to invent a section, a case, or a rule.
- Offline-capable. The index runs locally. No external API is called at query time.
- No data retention. Query text is redacted of PII before logging. Nothing is stored on the query path.
- Auditable. Every retrieval is written to a hash-chained audit ledger. Any tamper breaks the chain.
Corpus
As of the current index, the corpus contains 12 source documents and 7,358 indexed chunks, drawn from the official gazette publications of the Government of India:
- Digital Personal Data Protection Act, 2023 (DPDPA)
- Bharatiya Nagarik Suraksha Sanhita, 2023
- The Indian Contract Act, 1872
- The Indian Evidence Act, 1872
- The Companies Act, 2013
- The Constitution of India
- Data Act (EU), Data Broker Registration Act
- Draft AI Bill
- Registration and settlement agreement forms
Who it is for
- Lawyers who need the exact text of a provision, cited, without a browser in the way.
- In-house counsel who need a private, offline tool for client-sensitive research.
- Compliance officers who need to reference statutes to the section level.
- Journalists and researchers who need to verify a claim against the primary source.
Who it is not for
- Anyone seeking legal advice. The engine does not give it.
- Anyone seeking an opinion on what the law should be. The engine only shows what it is.
- Anyone expecting a chatbot. There is no conversation, only retrieval.
Built under
The Advocacy — A Law Firm
Advocate Registration No. D-180-2017 · Bar Council of Delhi
www.advocacyalawfrim.in
upmanyu@advocacyalawfrim.in · +91 97186 65039