The Architecture
Prajna. Shunya. Medha.
In sequence. Compounding.
Each layer is independently consequential. Together, they form something that cannot be assembled from parts — because the integration is structural, not cosmetic.
How the layers compound
Prajna
Reads the environment
Makes Shunya's propositions credible — they emerge from a calibrated corpus, not from instinct applied to dashboards. The 90-day baseline takes months to build. The corpus depth Shunya requires takes years to accumulate.
Shunya
Generates original thinking
Makes Medha purposeful — the simulation tests evidence-grounded positions, not generic messaging. A simulation without provenance is theatre. A simulation testing a position that emerged from the actual environment is intelligence.
Medha
Simulates before deployment
Refines Prajna — comparing real-world propagation against simulation predictions is how the reading improves over time. The architecture is not static. It learns.
Patent pending. The 90-day baseline calibration takes months to build. The corpus depth that Shunya requires takes years of sustained deployment to accumulate. The agent calibration in Medha is only predictive because it draws from that corpus. Each layer compounds the others. The architecture is not something that can be assembled quickly — and that is the point.
01 · Read
Prajna
Monitors information environments in real time. 65 languages — including Chinese social media platforms that no other institution in India currently reaches. 15-minute cadence. Continuously.
What it reads is not sentiment. Positive/negative binary is insufficient for any decision that matters. Prajna maps emotion across fifteen calibrated registers across three terrain types.
Resistance terrain
What will block or reverse your communication before it lands.
Receptivity terrain
Where your communication will be carried and amplified.
Liminal terrain
What is undecided — persuadable, but differently.
The discipline
A surge without a 90-day baseline is noise. A surge against a calibrated 90-day norm is a finding. This is what separates reading from guessing. No pattern is declared anomalous without that calibration.
Pride & Nostalgia — cross-terrain registers
Two registers that sit across terrains depending on how they are handled. Pride correctly acknowledged: the most powerful amplification vector available. Misread: flips to Humiliation. This is the hardest outcome to predict from desk analysis.
Narrative lifecycle tracking
Emerging → Accelerating → Plateauing → Receding. Quantified against 90-day baselines. Coordination detection: identifies when content moves with structural uniformity that organic discourse cannot explain.
02 · Think
Shunya
Monitoring tells you what the environment looks like. It stops there. Shunya does what monitoring cannot: it reads the subject terrain and tells you what it means — and what you can credibly land on it.
From the corpus Prajna has built, Shunya generates 5–6 strategic propositions. Not data summaries. Not dashboards. Actionable positions that emerge from the corpus itself — each evidence-chained, each provenance-tagged so every claim can be followed back to its source.
The analytical systems beneath Shunya were built over years of research into how information environments are structured, how narratives propagate, and what distinguishes positions that travel from those that fail in transit.
Five routes to an original position
Departs from the established interpretation
Names a boundary condition the prevailing view has missed
Changes the category in which the problem is understood
Carries an existing framework into territory it has not reached
Describes a phenomenon no existing frame has addressed
03 · Simulate
Medha
200–500 AI agent personas — journalists calibrated by outlet and beat, politicians by affiliation and record, activists by cause and register, influencers by platform and audience, general public by demographic and community.
Each persona has distinct profiles, reaction patterns, and influence radii, built against real emotional baselines from the Prajna corpus. These are not generic archetypes. They are modelled on the actual environments the intent will enter.
Feed in the proposition. Watch it propagate over 48–72 simulated hours. See where it lands, where it fails, what mutations it undergoes, which environments receive it and which reverse it. Revise. Test again.
The Pride Flip
The single hardest outcome to predict from desk analysis. The most consequential when it occurs. Pride correctly acknowledged is the most powerful amplification vector available. Misread — particularly when a proposition activates nostalgia in an aspirational register — flips to Humiliation. Medha is built to catch this before it enters the real world.
Sovereignty
Fully sovereign deployment. Zero cloud dependencies. No data leaves the premises. Available for Defence and any client where data sovereignty is non-negotiable.
The version that enters the real world has already been tested
against a model of the world it was designed for.