Pramana (working name) by VedAGI · Verified AI for life sciences
AI can draft the pack. It cannot be the source of the figures.
Today, Pramana recomputes each figure from its source, labels every estimate, and locks the figures until a named person signs. The receiver gets a signed receipt they can check in a browser, without installing anything. Being built, not live yet: each figure will become a chain of locked versions, each signed by a named person, where version two cites version one.
Products
Pramana (working name) is the lead product, shown above. Serlyx is the second.
For market access, HEOR (health economics and outcomes research) and evidence teams. It recomputes each figure from your own model, labels every estimate, and locks the figures until a named person signs. The receiver gets a signed receipt they can check in a browser.
Being built, not live yet: a chain of locked versions for each figure.
For underwriters. Serlyx helps them assess AI risk and E&O (errors and omissions) exposure for companies that build or use AI. It produces a structured risk assessment and a binder-ready summary. Every number it cannot verify from source is labeled as an estimate, and a human underwriter makes the decision.
Who they are for
Pramana (working name): market access, health economics and outcomes research (HEOR), and evidence teams who already produce payer and health technology assessment (HTA) packs, including AMCP (Academy of Managed Care Pharmacy) dossiers, and the person who has to sign them.
Serlyx: E&O and technology-liability underwriters at insurers and MGAs (managing general agents) who need to assess the AI risk of companies that build or use AI.
What Pramana (working name) does today
A per member per month (PMPM) cost is a value with a unit and a kind, not a clause in a paragraph.
Formulas run in code. Models may propose a formula. They do not write the value.
Unlocked or unlabeled figures cannot export.
What was computed, from which sources, under which formula, signed by whom.
Being built, not live yet: each figure will become a chain of locked versions. Each version will be locked on its own by a named person, and a revision or clarification released through Pramana (working name) will be a new version of a figure it already holds, with a receipt that cites version one. A payer's adjusted scenario will stay separate from the manufacturer's signed version.
What Pramana (working name) is not
Prose can be drafted anywhere, including by AI. Pramana locks the figures the prose points at.
You keep your model. It stays the calculator, and the reviewer still opens it. Pramana recomputes from it.
A human with the signer role locks the figures and stays accountable for the release.
Why VedAGI
AI is moving fast into life sciences. Large language models (LLMs) can produce figures that look right but cannot be traced to a source. In a payer or HTA pack, that is a risk no signer should have to carry.
Our answer in Pramana (working name): AI may plan the work. Code computes the numbers. A named human signs and stays accountable.
The same idea guides Serlyx: the underwriter, not the model, makes the decision, and estimates are labeled as estimates.
Human-aligned
Human-aligned, for us, means the person who carries a high-stakes decision stays accountable. AI strengthens that person's judgment and never replaces it.
In Pramana (working name), a named person with the signer role locks the figures and answers for the release. In Serlyx, the underwriter makes the decision. AI supports that person. It does not sign or decide.
In Pramana (working name), models may plan the work and propose a formula. Code computes the value from your sources. A model never writes the final number.
In Pramana (working name), each figure records its sources, its formula, and who signed it, so payers, HTA reviewers, auditors, and boards can check it.
In both products, an estimate is labeled as an estimate. In Pramana (working name), when the evidence is thin, the figure is marked ESTIMATE in the workbench and in the exported table, so it does not pass as computed.
Your data stays on your side. In Pramana (working name), the receipt carries source fingerprints, not your files. Serlyx runs locally and works offline.
Industries
Pramana (working name) is built for life sciences first: the market access, HEOR, and evidence teams whose figures go to payers and HTA reviewers. Serlyx is for insurance underwriting.
If you are about to send a table you cannot recompute, talk to us.
If you assess the AI risk of companies you insure, talk to us about Serlyx.
VedAGI. Intelligence for Humanity.