KALAP

The problem

Campaigns are run on gut feel and stale data. By the time a field report reaches the candidate, the moment to act on it is often already gone. Written tallies decay fast — accuracy drops sharply after just a day of relying on memory alone. Sentiment, community issues, and opponent activity live scattered across separate notebooks, spreadsheets, and verbal reports, so nobody holds the whole picture. The result is that risks surface only after the damage is done: a rumor or a rival's ground push reaches leadership after it has already shifted a barangay, not before.

Key challenges

Turning unstructured field conversations into decision-ready intelligence at scale is the core problem. Interviews are natural and unscripted by design — no forms in front of the constituent — which means the structure has to be recovered afterward by AI, not captured up front. The same platform has to roll up cleanly for a national or senatorial race and drill all the way down to a single barangay for a local one. And because this is constituent data, privacy can't be bolted on — it has to be designed in, with anonymized aggregation by default and gated access to raw detail.

What we built

KALAP is a case of rebuilding the workflow around agents rather than bolting a chatbot onto the old one. It runs a three-step loop — Collect, Categorize, Act. Field agents hold natural, conversational interviews with constituents, with no scripts or forms in the way. An agentic AI pipeline built on Anthropic's Claude then transcribes and sorts every conversation into issues, sentiment, aspirations, and candidate perception. The campaign team sees ranked priorities and suggested actions on one live dashboard, by barangay, city, or region — refreshed continuously, not once a week. The executive dashboard tracks candidate and opponent perception (positives, negatives, and favorability trend for every named figure), the top community-level and household issues ranked and mapped by barangay, constituent aspirations in their own words for messaging, and the criteria voters actually decide on. An agentic AI analyst — built on Claude — sits inside it, so leadership can ask directly, "Which barangay needs attention this week?" — and automated weekly reports deliver sentiment digests and perception summaries. It was built for privacy from day one, with data-protection compliance in mind, anonymized aggregation by default, and gated access to raw detail.

Our approach

1. 1

   Collect — natural conversational interviews

   Field agents interview constituents conversationally, with no scripts or forms in front of the person. The structure is recovered later by AI, so the interview stays human and the data stays honest.

2. 2

   Categorize — AI turns talk into structure

   Every conversation is transcribed and sorted into issues, sentiment, aspirations, and candidate perception. The messy reality of a real conversation becomes a queryable, continuously-refreshed intelligence layer.

3. 3

   Act — one live executive dashboard

   Ranked priorities and suggested actions surface by barangay, city, or region. Every card drills down for detail, and an embedded AI analyst answers direct questions like which area needs attention this week.

4. 4

   Roll up nationally, drill down locally

   The same platform aggregates sentiment across a region for a national or senatorial race and drills to full barangay granularity for a mayoral or local one — one system, every scale of race.

Key architectural decisions

Results

• Field conversations converted into a live, AI-categorized intelligence layer
• Continuous refresh — sentiment and issues tracked as trends, not weekly snapshots
• Executive dashboard: candidate/opponent perception, area and household issues, aspirations
• Embedded AI analyst for direct questions and automated weekly reports
• Barangay-to-region roll-up and drill-down on a single platform
• Issues ranked and mapped to the barangay level
• Privacy-by-design — anonymized aggregation by default, gated access to raw detail

Impact

KALAP replaces gut feel and stale tallies with a continuously-refreshed picture of the ground — what people are worried about, how a candidate and their opponents are perceived, and where the trend is moving, mapped down to the barangay. Because ground intelligence compounds, the earlier it's in the field the more a campaign is steered by trend rather than a last-minute snapshot. As an Anthropic Claude partner, AR Data built it the way it builds every agentic system — the workflow redesigned around agents and shipped as a production system end to end, on government-grade engineering discipline and privacy-by-design — turning the scattered reality of field reporting into a single intelligence layer campaign leadership can actually act on.

Tech stack