Sovereign deep tech for documents that can’t afford a wrong answer
shodasa is a deterministic reasoning engine, built in India, that reads formal documents and returns what they require — with the source clause attached. It runs inside the customer’s own environment, and nothing leaves it.
What it is, and why it matters
What. shodasa reads formal English documents — Acts, contracts, SOPs, tenders, policies — and builds a map of who must, must not and may do what. Users ask questions in plain English, and every answer points to the exact clause it came from.
Why. In regulated work, a confident wrong answer costs far more than a slow right one. An AI chat predicts text: its answers can change from one ask to the next, it can state something the document never said, and it spends tokens on every document. shodasa reads the document’s own sentences, so the same question gives the same answer, with its clause and an audit trail, at a significantly lesser cost.
The deep tech. The engine is not a language model. It is a deterministic reasoning engine, and the method is the subject of a filed Indian provisional patent.
Who it serves. Compliance and regulatory, legal and contract, quality and audit, finance and risk, bid and tender, and training teams — anyone who must know exactly what a document requires and prove it.
Evidence
Sample benchmark charts from real documents, followed by pilot results.
Results so far
- Legal documents pilot: Indian legal Acts — shodasa scored 85% accuracy.
- Legal judgements pilot: finding the precedents quoted in Indian court judgments — shodasa scored more than 75% accuracy.
- Engineering documents pilot: finding procedures (how to remove a heat sink) and reasons for alarms (why the NIC indicator is green) — shodasa scored more than 70% accuracy.
- Compliance documents pilot: Pharma eCFR 211 obligations analysed to find why a warning letter was issued and what response times are specified.
These are pilot results, not production deployments. You can test shodasa yourself on the live demo with your own document.
Sovereign and private by design
Many organisations — in defence, government, energy, pharma and finance — cannot send their documents to a cloud AI service. shodasa runs inside the customer’s own environment and has been tested running fully air-gapped, with no internet connection required, so documents and questions never leave. The technology is developed and patent-filed in India.
The idea behind it
About 2,500 years ago, Pāṇini wrote a rule system for Sanskrit that gives every participant in an action a defined role — who does it, to what, with what, and where. shodasa runs on PANINI-NET, a neuro-symbolic, deterministic engine inspired by Sanskrit grammar, which applies that structure to formal English. It reads who must do what from the grammar of the sentence itself, rather than predicting text.
Where things stand
shodasa is at prototype stage. It reads English documents in text form (text-based PDF or TXT); scanned images are not yet supported. We would rather be clear about that than overpromise, because an audit trail is only worth something if it is honest about what it covers.
We are open to beta testing. If you have documents and the questions your team asks of them, we would like to run them and show you what comes back.