अन्नदृष्टि Pitch Runbook

The problem, in their words

700+ LMT in the central pool at peak. 2,000+ depots. 80 crore NFSA beneficiaries. The network breathes in two violent waves — RMS wheat April to June, KMS rice October to January — and PDS offtake drains it every single day.

The data already exists. DOS, data.gov.in, NFSA, Annavitran. What does not exist is one picture that puts space, stock and time together for the officer on duty. So overflow gets discovered after the rakes are queued, and a stockout after the district has already run dry.

Don't say "digital twin" yet. Earn it.

The game board — one feeling: this is our network, and it is alive

1. Open on the national view. Say nothing for five seconds — let everyone find their own state.
2. Play the year. Run the wheat wave April → June. Point at the columns turning red.

   Amber is normal. Red is above 90 percent. Cyan is depleting. Nobody trained you on that legend — you already read it.

3. Pause on the late-April overflow.

   Nothing told the model to do this. Real procurement calendar, real capacities, and this falls out.

4. Tap the 🚆 रेल tab — everything else drops away. One tap isolates a layer. You can see what a rake is carrying from its wagons.
5. Drill state → districts → depot registry.

   Geometry is real. The UP depot registry is transcribed from the FCI board itself. The daily operational series in this build is synthetic and seasonally shaped — and I'll show you exactly which layer is which in a moment.

Volunteer the real/synthetic split before anyone asks. That's what makes the rest credible.

The decision, not the picture — the twin is the hook, the queue is the product

1. Rake Optimizer. Show the controls: 280 weekly rake quota, 28-day horizon, 21-day buffer, P90 confidence.
2. Turn the constraints on one at a time — protect rain-exposed CAP stock, enforce the 12-day destination floor, freeze movements inside 48 hours.

   These are your rules. The optimizer is not allowed to be clever about them.

3. Run rolling-horizon plan. Let the queue populate.
4. Walk the top recommendation end to end: priority lane, need, rake, ETA P50/P90, reason composition.

   Every row carries its own evidence — utilisation, coverage days, serving depot. Nothing arrives as "the system says so."

5. Press Why, read the reason composition aloud, then Approve.
6. The beat that matters: take another row and Override. Show it landing in the audit trail.

   The engine detects and suggests. The officer decides and logs. Override is a first-class action, not an escape hatch — and it's recorded, so next quarter you can see where the model was wrong.

7. District Allocation → pick a destination state → run.

   Rakes granted to a state are broken into district lots at the entry railhead. Pass one lifts every district under 7 days. Pass two brings everyone to the buffer floor. Pass three spreads the residue pro-rata to daily offtake.

8. If the room is engaged — Scenario Laboratory: early heavy monsoon + 20% rake shortage, 500 paths.

   P50 is the plan you execute. P90 is the plan you survive.

What is real, what is not — don't demo this, read the table

Boundaries: Survey of India, LGD. Depot registry: fci.gov.in annexures, DOS, Sansad Q&A. Daily stock: synthetic today, data.gov.in FCI datasets as the drop-in. Offtake: DFPD bulletin, NFSA, Annavitran, iMPDS. Movement: stylised corridors today, FOIS by partnership.

Every row in the shipped database is provenance-tagged — per-row confidence, per-layer source. When a number is a guess, the system says so.

This is the slide a technical evaluator remembers. Spend the full three minutes. Never cut it.

Path to production

Registry truth (4 wks) → live time axis (4 wks) → PDS coverage (6 wks) → movement (ongoing).

On-prem or NIC-cloud. Single-file web client. No external dependency at runtime. Nothing in this demonstration needed the internet, including right now.

The ask, then stop talking

A 12-week paid pilot with one Regional Office. We propose RO Lucknow, because the UP registry is already in the build.

Scope: live registry, the daily stock feed, duty-mode rollout to the AGM and D.O. desks. Success measured on two numbers — decision latency, and overflow-days.

We need three things: DOS read credentials or daily extracts, one nodal officer, and the RO wall-board data as ground truth.

अन्न का हिसाब, आँख के सामने। The grain ledger, made visible.