Tejas
तेजस्Deterministic governance · a P9e Microsystems product
Tejas is the answer to a problem that arrives the moment AI stops answering questions and starts operating things. A model that gives a different answer to the same input cannot be certified, cannot be replayed, and cannot be defended when someone asks why it did what it did. Tejas does not make the model deterministic — it wraps the execution in a layer that is, so the run can be reproduced, proven and explained afterwards.
- Type
- Embeddable layer
- Layers
- Tejas · Dhruva · Siddhanta · Dristi
- Runs
- Edge, on-prem, cloud, offline
Confidence is not correctness, and a log is not a causal trace. Anything that controls physical infrastructure needs an execution record you can reproduce exactly and prove cryptographically.
How it fits together
Four layers, four guarantees
Each layer owns one property and can be used on its own. The bottom row is the part most architecture diagrams leave out: what you actually lose by omitting a layer.
What each layer owns
Each works independently and improves the picture on its own. Together they close the chain from execution to a certificate an auditor will accept.
Tejas
Deterministic execution governanceThe compiler and runtime layer. Converts a pipeline into bounded deterministic execution and emits an execution manifest, a capability manifest and a graph fingerprint. Without it, behaviour is structurally unpredictable.
Dhruva
Deterministic state controlFreezes and fingerprints execution state — model weights, environment, runtime snapshot. Without it, replay is unreliable, because the same graph on a drifted environment is not the same run.
Siddhanta
Cryptographic proof & attestationSigns, verifies and anchors the run. Takes the state hash, graph hash and artifact hash and issues an execution certificate with a signed ledger entry. Without it, governance is unverifiable.
Dristi
Deterministic renderingMaterialises the output the same way every time and hashes the artifact. Without it, the thing a regulator or a court actually looks at is legally disputable.
Praman
Deterministic precision systemThe computational engine — speech, vision, spatial, temporal and symbolic reasoning in one core, with plug-in domain modules. One codebase serving agriculture, healthcare, robotics and defence rather than four forks of it.
What Tejas does
Bounded execution, not best-effort
The pipeline is compiled into an execution graph with declared bounds. Structural, authority and execution determinism are enforced before and during the run, not asserted afterwards.
Exact replay
State, weights and environment are fingerprinted together, so a run from six months ago can be reproduced byte for byte rather than approximately re-created.
Cryptographic attestation
Every run produces a signed execution certificate anchored to a ledger entry. The claim 'this is what the system did' becomes something you can hand to an auditor.
Modular, and it degrades honestly
The layers work independently — each one improves the picture on its own, and the documentation says plainly what you lose by leaving one out. Runs offline, online or hybrid, from edge devices to enterprise clusters.
Who Tejas is built for
Not a list of industry names — what the product actually does differently in each.
Aerospace & unmanned systems
Where an autonomy decision has to be reconstructed after a flight, from evidence rather than recollection.
Industrial & robotics
Machine vision and control decisions on a line, where a rejected batch becomes a dispute unless the run can be replayed exactly.
Agriculture
Advisories a grower acts on and an insurer or a regulator may later question. Reproducibility is what makes the advice defensible.
Regulated sectors generally
Healthcare and defence are the obvious cases, and the ones we are least far along in. We are naming them as intent, not as a deployment record.
Deployed and embedded
- Vyovrinda — flight autonomy and mission decision paths
- Sasyaveda — agronomic advisories that have to be defensible
- Industrial Automation — machine vision and control decisions
- Third-party systems, embedded via the SDK
Where it actually stands
In pilot deployment. The architecture and the layer contracts are settled; the surrounding ecosystem — certifications, third-party audits, regulatory recognition — is not, and neither is anyone else's in this field yet.
We state maturity on every product page. A claim we cannot defend costs more than the work it wins.
Talk to us about Tejas.
Deployment, embedding, evaluation licences — tell us the system it has to live inside.