We make physical reality computable.
Bhārat first. The world next.
Building the verifiable grounding layer for physical-world AI. One compounding system: understand a place, run a mission on real hardware, and prove what happened.
Enter the thesisसं Saṃsāra — the understanding layer
A locality grammar that keeps AI grounded.
Saṃsāra turns demographic, environmental, infrastructure, land-use and geospatial signals into a witnessed, machine-readable model of place — Bhārat Strata. It gives models and agents a factual grammar of locality to reason over, so they answer from evidence instead of hallucinating the physical world.
Bhārata Strata · versioned locality knowledge model · 264,583 cells · 492 columns
The loop
संSaṃsāra understands a place. · कKarma acts on it and proves the outcome. · Every proof sharpens संSaṃsāra.
क Karma · the action layer
Turn understanding into action — and prove it.
Karma is action, and its proven consequence. It acts on what Saṃsāra knows — enroll any device, run a mission with on-device ML at the edge, and return signed, tamper-evident proof of what happened.
Its first commercial expression is programmatic out-of-home advertising — planned by location intelligence, run on real screens, billed on proof-of-play. The way Google and Meta made digital programmatic, brought to the physical world.
Run
The device that runs the mission at the edge, acts on its own, and signs every observation.
Mission EdgepDOOH
Programmatic out-of-home — planned by location intelligence, run on real screens, billed on proof-of-play.
GGOS · mission control for autonomous hardware · multi-tenant fleet · signed, tamper-evident evidence
The gap between intention and reality
The physical world struggles to answer the questions software made trivial.
A campaign runs, but was it seen? An asset is installed, but is it working? A report is filed, but does it reflect reality — or just ask for trust?
- 01What actually happened?
- 02Where did it happen?
- 03To whom did it happen?
- 04Under what conditions?
- 05Did the intended outcome occur?
- 06What should happen next?
GaliGali exists to answer them with evidence — by witnessing reality directly and holding every claim to a standard of proof.
In practice · out-of-home, end to end
From the right audience to proof it landed.
A brand wants to reach a certain kind of person across the city. The screens are already everywhere — billboards, metro panels, mall and retail displays — run by dozens of different operators. Lighting them up was never the hard part. Knowing which ones matter, showing the right thing to whoever’s actually in front of them, and proving it landed — that is.
The mission
Put the brand in front of its most likely audience — the right neighbourhoods, the right moments — and prove every impression it earns.
- 01Saṃsāra · GridRock
Understand
GridRock reads the city the way a planner only wishes they could — who lives where, how they move, when a street fills up — and finds where this brand’s people really are.
- 02where they’ll see it
Target
From that, it picks the screens, streets and times of day that will actually reach them — not just the cheapest slots going.
- 03Karma · GGOS
Act
GGOS takes over any screen — whoever owns it — sees who is actually standing in front of the glass, and plays the creative that fits the moment.
- 04signed evidence
Prove
Every play is signed the instant it happens — where, when, to how many — into proof an advertiser can check for themselves, not a number to take on faith.
- 05back to Saṃsāra
Compound
What actually happened on the street feeds back into GridRock — so the next campaign starts smarter than the last.
↺ the more it runs, the better it knows where to aim
The canvas · the length & breadth of out-of-home
Objective met — real reach you can check, not impressions you have to trust. And the system knows the city a little better for next time.
In practice · Saṃsāra · public water
Every monsoon it fills. Every summer, it vanishes.
Across Bhandara district in eastern Maharashtra, the old malguzari tanks brim to reservoir scale each monsoon, then drain to bare ground by summer — the water lost to silted, shallow beds. New check dams and percolation tanks could hold far more. But which sites, out of thousands — without displacing anyone, or funding a pond that won’t hold? Usually guesswork and pressure decide.
And it isn’t one tank. Across the district, ~1,755 ha of tank water at the monsoon peak drains to about 46 ha by April — ≈2,394 football fields of water, gone every summer.
The approach · terrain to sanction
So we let the land, the sky and the map decide.
It is Saṃsāra — the same engine that reads a city to find a brand its audience — pointed at a district instead of a demographic. No campaign at the end of it; only water that has to last.
The task
Find where Bhandara should build and desilt next — sited on the real drainage network, proven from orbit, honest about who it moves — and quantify the net-new water each one buys.
- 01GLO-30 · WhiteboxTools
Read the terrain
A 30 m elevation model, depression-breached, becomes the district’s full drainage network — every nala ordered, the Wainganga at the top.
- 02AHP · 7 criteria
Site & score
Slope, soil, land use and catchment gate what can be built where; a Saaty-weighted score ranks every candidate on the real network — nothing off-channel.
- 03Sentinel-2 + Sentinel-1
Witness from orbit
Four monsoons of optical and radar water extent cross-witness every fill and draw-down — radar sees through the cloud that blinds optical when tanks fill.
- 04Overture · 415,175 buildings
Screen for people
Each design footprint is checked against real buildings; bunds step down to clear settlements, and tank renovations displace no one.
- 05tiers · delta impact
Rank & deliver
Sites resolve into a shortlist, a worth-surveying tier and the rest — each with the net-new water it would add, counted against its own history.
↺ every step a witnessed artefact — reproducible end to end from scripts/water/
57 / 163,000
sites chosen from the district’s stream cells — nothing off-channel
415,175
real building footprints screened; the tank revivals displace no one
r = 0.9
radar and optical, opposite physics, agree this tightly on the flagship reservoir
4 bunds
shrunk from 3 m to 2 m to spare homes — and they still fill
The outcome · what it delivers
0.0MCM
of net-new water a year — a 25-site programme, counted incrementally over what each site already holds
- ~522k
- drinking person-yrs
- 5,743
- ha irrigable
- 24
- villages
Objective met, and honest — one tank alone taps a catchment larger than Mumbai (757 km²), every hectare acquired returns ~69 TCM a year, and on black-cotton clay only 12% is booked as recharge; the rest is surface storage, not overclaim.
The operating model
Two divisions, one reality.
Saṃsāra reads reality and Karma acts on it. What actually happens is witnessed and signed, and that ground truth flows back to sharpen what Saṃsāra knows — a loop that compounds with every turn.
Saṃsāra
know the world · read
GridRock reads the one reality at scale — satellite · census · geospatial · public records.
Karma
act in the world · read + write
GGOS — the operating system every mission runs on.
Operate · Mission Control
Govern the whole fleet from one console — enroll, deploy, command.
Run · Mission Edge · the node
Runs your apps at the edge, acts on its own, and signs every observation.
Extensible hardware
Billboards · TVs · screens · drones · cameras · IIoT · robotics.
REALITY
the one world — Saṃsāra reads it · Karma acts on it
Signed Evidence
reality, verified · e.g. pDOOH
Witnessed, not modeled
Built on the world’s authoritative open data.
Nothing here is synthetic. Every dataset below is one we actually ingest, version and witness — fused into Bhārata Strata and our products, held to a falsifier suite designed to make it fail, and signed.
Earth observation & climate
Built environment & maps
Population, water & air
India — admin & open data
Aviation & space situational
Contains modified Copernicus data · ESA WorldCover, WorldPop, GHSL & EDGAR (EC JRC) and geoBoundaries under CC-BY · NASA (NASADEM, HLS, IMERG, GEDI) & ASF DAAC · NOAA · Maxar Open Data · CHIRPS (UCSB / USGS) · Google Open Buildings, Microsoft Building Footprints and Overture Maps · © OpenStreetMap contributors (ODbL) · OpenAerialMap · Mapillary · NRSC / Bhuvan & MOSDAC (ISRO) · India-WRIS, CWC, IMD, CPCB, NHAI, MoRTH Vahan, and data.gov.in (LGD, Census 2011, Udyam) — Government of India · ERA5 / Copernicus C3S (ECMWF) · OpenSky Network · CelesTrak · OurAirports · Wikidata. Logos are the property of their respective owners and shown for source attribution only.
The thesis, in one flight
From the witnessed substrate over India, down to a single screen, and back.
One continuous flight: observe a place, decide where to act, execute on a real edge device, verify what happened — and watch the loop compound across the country.