Cluster C · Field-tech stack

AI for Booth Workers, Panna Pramukhs and ICT Pramukhs: The Field-Tech Stack

How AI voice agents integrate with India's booth-level karyakarta network — panna pramukh apps, ICT pramukh dashboards, real-time field intelligence, and the human-AI division of labour.

8 min readUpdated 22 May 20261,638 words

India's electoral campaigns are won at the booth level. A single Lok Sabha constituency has 5,000-25,000 polling booths. Each booth is staffed by 1-2 karyakartas from each contesting party. The karyakarta's job is to know the 800-1500 voters in their booth, persuade the undecided ones, and turn out the supportive ones.

This guide is about how AI voice agents integrate with this booth-level field-tech apparatus. The role hierarchy, the technology stack today, the integration patterns that work, and what's coming for 2027-2029.

The booth-level hierarchy

Different parties use different terminology, but the role structure is broadly similar.

BJP's hierarchy (panna pramukh model)

  • Panna pramukh: assigned to one page (panna) of the voter list — typically 60 voters in close geographic proximity. Lowest field-worker level.
  • Booth pramukh: in charge of the entire polling booth — typically 800-1500 voters, with 14-25 panna pramukhs reporting in.
  • ICT pramukh (ward/sector level): coordinates 5-15 booths.
  • Mandal pramukh: coordinates 50-100 booths at the mandal (sub-district) level.
  • Vidhan Sabha incharge: coordinates the entire constituency.
  • Lok Sabha incharge: coordinates the entire parliamentary constituency.

The panna pramukh is the genuine innovation. Pioneered at scale in Gujarat in the early 2010s and rolled out nationally, it sub-divides the booth-level workload into manageable pieces. 60 voters is a number one person can genuinely know — neighbours, families, individual concerns.

INC + regional parties

  • Booth-level worker (BLW): equivalent of booth pramukh
  • Polling agent: on the day, sits at the polling booth representing the candidate
  • Block-level coordinator: roughly equivalent of mandal pramukh
  • District president: constituency-level coordinator

Smaller granularity (no panna pramukh equivalent in most regional parties) means each booth-level worker has to handle 800-1500 voters alone — operationally harder.

Common denominator

Regardless of party, the role structure compresses voters → ~60-1500 voters per ground-level worker → coordinators above them → constituency-level command.

The technology challenge: keep every level synchronised on current voter status, current outreach activity, current sentiment, current polling-day progress.

What field-tech looks like today

A survey of what booth workers actually use in 2026.

BJP

  • NaMo app: outward-facing app for supporters
  • Saral app: booth-level operations (panna pramukh data entry, voter list access)
  • Mission 2024 app: campaign-specific tool used in 2024 cycle
  • Internal dashboards for booth pramukhs and above

Most disciplined data infrastructure of any Indian political party. Strong WhatsApp group ecosystem on top of the apps.

INC

  • INDIA app variants: state-specific apps with varied capability
  • Custom CRMs: state-level deployments differ
  • WhatsApp-heavy coordination

Less centralised than BJP. Quality varies sharply by state and by election cycle.

Regional parties

  • DMK (Tamil Nadu): strong booth-level digital coordination, custom apps
  • TMC (West Bengal): heavy WhatsApp-driven workflows, less app-centric
  • AAP: custom apps for Delhi/Punjab cycles, less developed elsewhere
  • BJD (Odisha): custom internal CRM
  • BRS (Telangana): mix of WhatsApp and custom apps
  • SP, RJD, JD(U) etc: mostly WhatsApp-driven with some custom data tools

The pattern: party-specific app for core field operations + WhatsApp groups for everything else.

The integration problem

The booth worker has their data. The AI voice agent has its data. The war-room dashboard has third data. Without synchronisation, they work at cross-purposes:

  • The panna pramukh visits a voter Tuesday morning; the AI calls the same voter Tuesday evening with a duplicate message
  • The AI captures a grievance; nothing flows to the local karyakarta who could solve it
  • The booth pramukh has a polling-day plan; the war-room has different priorities; the AI is operating on a third schedule

The cost of this desync: 20-40% reduction in cumulative campaign effectiveness. Every duplicate touch wastes budget; every missed handoff loses a vote.

The integration pattern

What good looks like.

A single voter record, multiple surfaces

Every voter has one record. The record contains:

  • Identity (voter ID, name, demographic, booth)
  • Outreach history (every voice call, WhatsApp message, SMS, door-to-door visit)
  • Current sentiment (latest classification + history)
  • Grievance tickets (open and closed)
  • GOTV status (have they been confirmed? have they shown up?)
  • Polling-day flags (need transport? need escort?)

Multiple apps read/write from this single record. The booth pramukh's app shows it. The AI platform writes to it. The war-room dashboard aggregates from it.

Real-time bidirectional sync

  • AI completes a call → updates voter record → karyakarta app reflects the update within 60 seconds
  • Karyakarta visits a voter → notes it in their app → AI platform sees the update before the next call wave
  • War-room marks a voter as "do not contact" → both AI and karyakarta see this within seconds

The technology is straightforward (event-driven backend, real-time push to mobile clients). The execution is hard because most party apps weren't designed with bidirectional sync in mind.

Workflow handoffs

Specific moments where AI and karyakarta hand off:

  • AI identifies a grievance → ticket created → assigned to local karyakarta → karyakarta closes the loop in person → karyakarta marks closed → AI follows up 30 days later to confirm
  • AI completes a GOTV call → voter expressed need for transport → ticket to karyakarta → karyakarta arranges → karyakarta confirms voter at booth
  • Karyakarta marks voter "strong supporter" after home visit → AI agent's future calls to this voter use a more relationship-focused tone

These handoffs are the heart of the integration. They're also where most platforms fail.

What the karyakarta sees on a mobile dashboard

A well-designed booth-level dashboard:

Home screen:

  • Their booth
  • Today's date and polling phase
  • Sentiment summary (positive / undecided / negative counts)
  • 5 top priorities for today

Voter list view:

  • 60-1500 voters in their booth
  • Filter by sentiment, last touch, has-grievance, GOTV status
  • One-tap to call (AI agent handles routing)
  • One-tap to mark visited

Polling-day view:

  • Live turnout count from ECI
  • Supportive voters who haven't yet voted (highlighted)
  • Transport requests (with phone numbers)
  • One-tap status update ("voter X confirmed at booth")

Communication:

  • Direct line to ICT pramukh / mandal pramukh
  • War-room broadcast messages
  • Escalation flow for urgent issues

Crucially: every action the karyakarta takes is logged, every datum the AI captures is visible to the karyakarta. Two-way information flow with no gaps.

What war-room sees

The aggregate dashboard for the constituency-level commander:

  • Daily AI call volume + completion across the constituency
  • Sentiment trends by booth, demographic, dialect
  • Booths with anomalous metrics (low engagement, high negative, high non-completion)
  • Karyakarta activity heatmap (who's active, who's lagging)
  • Open grievance tickets count by booth
  • GOTV progress by booth (polling-day specifically)
  • Anomaly alerts and recommended actions

The war-room sees patterns the booth-level karyakartas can't see (cross-booth comparisons, emerging issues, infrastructure problems). The booth-level karyakartas see things the war-room can't see (specific voter relationships, local dynamics). The integration surface lets each level work with the right data.

Polling day: the integration peaks

The entire field-tech apparatus operates at maximum intensity for ~14 hours on polling day. The integration patterns above all converge.

A morning: AI calls supportive voters with their booth info. Voter responds "need transport". Karyakarta gets a notification. Karyakarta dispatches their tempo. AI follows up: "क्या आपकी transport आ गई?". Voter confirms. Karyakarta brings them to the booth at 11 AM. Karyakarta marks on the app "voter X at booth". ECI turnout app confirms by 12 PM. AI no longer calls voter X for the day.

That sequence — AI call → voter response → karyakarta action → bidirectional updates — happens hundreds of thousands of times per constituency on polling day. Every successful sequence is one supportive voter who definitely showed up.

Common failure modes

1. Apps don't talk to each other

The party's existing field app and the AI platform are separate systems with no integration. Karyakartas duplicate effort; data drifts. Common in 2024 cycle; expected to be partially resolved for 2027.

2. Latency in updates

Karyakarta visits a voter at 10 AM; AI calls the same voter at 11 AM because the sync took 90 minutes. Voter feels harassed. Real-time sync (under 60 seconds) is the bar; daily-batch sync is not enough.

3. Over-reliance on one or the other

Some campaigns deploy AI without strengthening the booth-level karyakarta network — the AI generates leads that nobody can close in person. Others have strong karyakartas but no AI — leaving the broadcast workload as a bottleneck.

4. Compliance gaps in karyakarta apps

The party's field app may not have DPDP-compliant data flows. When integrated with the AI platform, the compliance weakness becomes the platform's weakness too. Audit the field app's data handling, not just the AI's.

5. Karyakarta training gaps

Senior karyakartas may resist app-driven workflows in favour of established patterns. Training and onboarding the field network on new tools takes weeks. Plan for this in the campaign timeline.

Where AiSewak fits

AiSewak's field-tech integration:

  • Real-time bidirectional API for party field-apps (BJP Saral, NaMo, INC variants, etc.)
  • Custom adapter framework for regional party apps (3-5 days to integrate a new app)
  • Native mobile dashboard (Android + iOS) for parties without an existing field app
  • Voter record single-source-of-truth design
  • Polling-day-specific real-time UI
  • DPDP-compliant data handling end-to-end

Where to go next

The booth-level karyakarta + AI voice agent integration is where elections will be decided in 2027 and 2029. Campaigns that figure out the workflow handoffs operate at a level of precision that purely-AI or purely-karyakarta operations can't match.

Frequently asked questions

What is a panna pramukh?

A panna pramukh is the BJP's micro-level field worker assigned to one page (panna) of the voter list — typically 60 voters living in close proximity. Pioneered at scale in Gujarat and Madhya Pradesh, the model is now used across India by multiple parties under different names.

Does AI replace the booth worker?

No. AI replaces the karyakarta's broadcast role (calling 60 voters with the same message) but cannot replace the relationship role (being the trusted face at the booth on polling day). The integration makes each karyakarta more effective, not redundant.

How do parties without panna pramukh equivalents use AI?

The architectural pattern works regardless of the party. INC's booth workers, regional parties' polling agents, AAP's volunteer networks — all benefit from the same dashboard + AI-call coordination model. The terminology varies; the operational pattern is the same.

What technology do booth workers actually use today?

Mostly mobile apps. BJP's NaMo app for outreach, Saral app for booth-level operations. INC has been building INDIA app variants. AAP uses custom internal apps. Most regional parties are on commercial CRMs or WhatsApp-driven workflows.

What's the biggest pain point in field-tech today?

Data sync. The booth worker has their own data, the war-room has different data, the AI platform has third data. Reconciling these in near-real-time during the campaign is the central challenge — and where AI-platform integration produces the largest gains.