Flipkart Gridlock 2.0 · Round 2 · Prototype

PWCA

Project Without a Cool Acronym

We don't just detect congestion —
we price it, then relieve it.

1 Find→ 2 Price→ 3 Predict→ 4 Dispatch→ 5 Prove

Decision-intelligence for parking-induced congestion · Bengaluru Traffic Police

→ next · F fullscreen · ? help

!

The Problem01 · THE GAP

A ₹500 fine hides a
city-scale cost.

Illegal parking steals road capacity → queues form → drivers idle → fuel burns → the air fouls. BTP logs every violation — the log never adds up where it hurts, what it costs, or where to send the next scarce officer.

Three ways to fight it. Only one scales:

A · Detect moreMore cameras✗ no imagery · detection ≠ relief

B · More bootsBlanket enforcement✗ officer-hours are scarce

C · Our approachPrice it, then optimise✓ 5% of spots = 61% of cost

382

City PPI

Parking-Pressure · Severe

₹13.68 L

Cost / day

≈ ₹49.9 cr a year

1,03,449

Veh-min lost / day

people stuck in the queue

3,135 kg

CO₂ / day

+ a PM2.5 health bill

1,698

Hotspots

across 54 police stations

298,445

Violations

Nov 2023 – Apr 2024

Source — BTP parking-violation log (anonymised), 151 days. Modelled estimates; every figure traces to a formula + cited parameter.

North-star

Vehicle-minutes of congestion relieved per officer-hour. Everything is measured against it.

The Platform02 · SOLUTION

One pipeline — from raw log to proven relief.

1

Find

298K violations → 1,698 hotspots, ranked by ₹ not counts.

2

Price

Attribution engine: violation → delay → ₹, fuel, CO₂, health.

3

Predict

Backtested forecast of tomorrow's pressure — beats baselines.

4

Dispatch

ROI optimiser + patrol route for scarce officer-hours.

5

Prove

Every clear booked to an auditable Impact Ledger.

PWCA command overview — **Fig. 01** — Command overview: the five-step story, in one screen.

1

Step 01 · Find03 · THE PIPELINE

Where the congestion
really is.

Every violation snaps onto an Uber-H3 grid and collapses into 1,698 ranked hotspots — sorted by the rupees of delay they cause, not by raw ticket count.

1,698

Hotspots

61%

of cost from top 5%

33

spots = half the cost

Hotspot map of Bengaluru — **Fig. 02** — Hotspots ranked by ₹/day, mapped over the city.

2

Step 02 · Price04 · THE PIPELINE

Turn each violation
into rupees.

The Congestion Attribution Engine runs the causal chain — and every number traces to a formula + a cited parameter in params.yaml.

Violation→ Capacity lost→ Vehicles delayed→ ₹ time · fuel · CO₂ · health

₹13.68 L

priced / day

3,135 kg

CO₂ / day

₹20.38 L

health / yr

City insights and cost concentration — **Fig. 03** — The cost is concentrated: fix 33 spots, halve the problem.

3

Step 03 · Predict05 · THE PIPELINE

Tomorrow's pressure,
today.

A gradient-boosted, time-series-backtested model ranks where pressure flares next. It's only credible because it beats the naïve baselines — and it does, with no leakage.

0.477

Precision @20

vs 0.385 / 0.335 naïve

+24%

lift over baseline

31

backtest days

Next-day forecast with baseline comparison — **Fig. 04** — Next-day forecast, scored against named baselines.

4

Step 04 · Dispatch06 · THE PIPELINE

Spend scarce officer-hours
where they pay.

An ROI optimiser — cost × risk, capacity-constrained — writes the work orders and the patrol route, then proves the lift over simply deploying to the busiest spots.

₹4.92 L

relief / day

≈ ₹17.96 cr / yr

+19.6%

vs busiest-spots

~26.5×

vs random deploy

Enforcement optimizer and work orders — **Fig. 05** — Work orders, ranked by relief — dispatched to the field.

5

Step 05 · Prove07 · THE PIPELINE

Every action,
measured & auditable.

Each clear is booked to an append-only Impact Ledger — rupees, vehicle-minutes and CO₂, traceable to the formula and the hotspot it came from. No vanity metrics.

Live

SQLite + SSE ledger

₹ · min · kg

booked per action

0-install

WhatsApp dispatch

Impact ledger with booked relief actions — **Fig. 06** — The Impact Ledger: relief booked, with provenance.

The Differentiator08 · TWIN RACE

Same officers. Same hours.
4.2× the relief.

One simulated day · 12 officers · 180 officer-hours. Random chases active hotspots; PWCA follows the optimiser. The relief diverges.

PWCA optimiser₹2,74,205

Random / by-count₹65,511

₹1,523

per officer-hr · PWCA

₹364

per officer-hr · random

The north-star, made visible.

Twin Race counterfactual result — **Fig. 07** — Two lanes, one day: cumulative cost (lower is better) splits apart.

What Changes09 · IMPACT

From reactive & blind to targeted, measured — and self-funding.

Reactive patrols → ROI-ranked work orders for the next shift.
A ₹500 fine → an itemised, auditable congestion bill.
Invisible cost → a live ledger of relief actually delivered.
Cost centre → recoverable municipal revenue.

Polluter-Pays · Sagar Theatre Junction

Its parking spillover costs the city ₹6.10 cr / year. 45% is recoverable as a congestion levy — ₹2.74 cr / year — alongside a Gemini-drafted official notice (with deterministic fallback).

Polluter-Pays itemised congestion bill — **Fig. 08** — An itemised ₹ bill: time, fuel, health, carbon → recoverable revenue.

Make it Tangible10 · SIMULATOR

Watch one parked car
choke an arterial.

A kinematic-wave digital twin: set how many vehicles, of what type, under what demand, and the queue backs up the road in real time at a modelled shockwave speed — magnitudes anchored to the Attribution Engine. Then drag TOW to clear the lane and book the relief.

₹1.12 L

priced / day here

62.7%

capacity lost

16 km/h

shockwave upstream

Butterfly-Effect kinematic-wave simulator — **Fig. 09** — The jam backs up in accelerated time; clearing relieves **₹1.03 L/day**.

Ease of Use11 · ASK & ACT

Type plain English.
Get a priced action — you confirm.

The docked copilot plans & drafts; you approve. Gemini parses intent + slots only — the server runs the trusted engines and owns every number. Nothing changes until you confirm.

Plain-English ask→ Gemini · intent + slots→ Trusted engine→ Priced card · you confirm

Anti-hallucination · works with no API key

“Send 6 officers to the top hotspots” → a dispatch plan relieving ₹4.53 L/day, +16.7% over chasing the busiest spots. Every figure from the optimiser; a deterministic parser is the floor when there's no key.

Ask & Act agentic copilot proposing a dispatch plan — **Fig. 10** — One box → a typed, confirmable card: plan · route · notice · navigate · answer.

The Public Layer12 · CITIZENS

Citizens report.
The city is held to account.

Drop a pin, pick what you saw — and the report is priced instantly by the same Attribution Engine that runs the war-room. The most-flagged spots surface straight into the enforcement queue — closing the loop between the public and the patrol.

Instant ₹

priced per report

6

report categories

→ queue

feeds enforcement

Citizen report — drop a pin, instantly priced — **Fig. 11** — Drop a pin → an instant ₹ price; the most-flagged spots feed enforcement.

Under the Hood → What's Next13 · CREDIBILITY & ROADMAP

Built to be audited — and ready to ship.

Now · Prototype

Decision platform

Find→Prove on historic BTP data
Twin Race + Polluter-Pays
Citizen reports + Copilot

Next · Field app

Officer mobile app

Live CRITICAL alerts
One-tap dispatch
"CLEARED" → ledger

Then · Live data

Real-time feeds

ANPR / sensor ingest
MapMyIndia network
Continuous re-optimise

Scale · Policy

City & citizens

BBMP / DULT dashboard
Levy as policy lever
Next cities

The stack · audit-ready

Python · pandas Uber H3 (res 9) XGBoost OR-Tools / PuLP FastAPI SQLite · PostGIS React · Vite Gemini + fallback

Every figure = formula + cited parameter; low / expected / high bands.
Time-series splits, named baselines, a model card per model — no leakage. Whole dataset (298K rows) processes in ~18s.

Works in 3 languages

Citizens & officers in their own language · runs with no API key (deterministic fallback).

What's Coming14 · FIELD APP · CONCEPT

The same engine — in the constable's hand.

01 Alert

14:325G

Critical alert

Elite Junction

Predicted top-1 hotspot, next hour.

PPI 382Severe

Building · est. ₹1,520/hr

View work order →

→

02 Act

14:345G

Work order · #PWCA-2381

Elite Junction

1.2 km · ETA 6 min

Relief if cleared · est. ₹4,200/day +7.8% vs busiest-spot deploy

Accept & navigate

Defer

→

03 Prove

14:415G

✓

Cleared

14:41 · booked to ledger

Relief booked+₹4,180

Vehicle-minutes+312

CO₂ avoided+1.7 kg

Provenance · Elite Junction · formula ✓

Proof, not vanity — same engine as the war-room.

Concept · design preview — figures illustrative. The shipped app books real ledger numbers from the same Attribution Engine + Impact Ledger that powers the war-room.

PWCA · Bengaluru

PWCA

Price the congestion. Relieve it. Prove it.

Find→ Price→ Predict→ Dispatch→ Prove

Flipkart Gridlock 2.0 · Round 2 · Prototype Phase

We don't just detect congestion —we price it, then relieve it.

A ₹500 fine hides acity-scale cost.

One pipeline — from raw log to proven relief.

Find

Price

Predict

Dispatch

Prove

Where the congestionreally is.

Turn each violationinto rupees.

Tomorrow's pressure,today.

Spend scarce officer-hourswhere they pay.

Every action,measured & auditable.

Same officers. Same hours.4.2× the relief.

From reactive & blind to targeted, measured — and self-funding.

Watch one parked carchoke an arterial.

Type plain English.Get a priced action — you confirm.

Citizens report.The city is held to account.

Built to be audited — and ready to ship.

Decision platform

Officer mobile app

Real-time feeds

City & citizens

The same engine — in the constable's hand.

Price the congestion. Relieve it. Prove it.

We don't just detect congestion —
we price it, then relieve it.

A ₹500 fine hides a
city-scale cost.

Where the congestion
really is.

Turn each violation
into rupees.

Tomorrow's pressure,
today.

Spend scarce officer-hours
where they pay.

Every action,
measured & auditable.

Same officers. Same hours.
4.2× the relief.

Watch one parked car
choke an arterial.

Type plain English.
Get a priced action — you confirm.

Citizens report.
The city is held to account.