Bakhabar Kisan · Call QA Platform

BKK QA

Upload agent–farmer call recordings in any audio format. Transcribes calls into Urdu (Roman Urdu on request), scores each call against SOPs you define, tags sentiment, and tracks agent performance over time.

Agent

SOP / scoring rubrics (tick one or more — the first ticked is the primary score)

Click to choose audio files or drag files / folders here — any format, one file or hundreds.

Processed

0 / 0

Avg score

—

Positive

—

Neutral

—

Negative

—

Results

Each call gets one full Urdu transcript (Roman Urdu view available per call), a rubric score with reasoning, sentiment, and a farmer profile.

No files processed yet.

Create a scoring rubric (SOP)

Define the criteria a call is judged against. Each gets scored 0–max by the QA model, with reasoning.

Rubric name

Criteria

Built-in SOP reference

The BKK “Daily Random Call Booster” QA sheet is available as a rubric with its exact scoring math (Behavior 60% as average of 1–3 scores; Resolution 40% from 15/10/20/5-point items; Fatal Error → 0, False pretext → ÷2, Performance Killer → ×0.75, Bonus +5, Behavior Booster ×1.07; grade A+ ≥96, A ≥91, B ≥86, C ≥80, else U). BKK Contact Center SOPs v2.0 (28 Jan 2025), the Inbound Profiling Script and the Livestock Queue Script are built into the judge. Every call is checked against S1–S14 below; the “BKK Contact Center SOP v2.0” rubric scores against them.

Show SOP rulebook

Existing rubrics

No rubrics yet.

Add an agent

Agent name

Agent performance

Aggregated from every call scored and saved on this device.

Agent	Calls	Avg score	Positive	Neutral	Negative
No data yet.

Call history

Filter by agent

Filter by sentiment

No calls yet.

Platform usage

Calls scored, average score, and sentiment mix across every call saved on this device (not just this session).

Total calls

Avg score

—

SOP misses

—

Conduct flags

—

Positive

—

Neutral

—

Negative

—

Farmer profile captured

—

Groq API usage today

Tracked client-side for this browser session only (resets on reload) against Groq's free-tier daily limits, so you can see how much runway is left. "Unconfirmed ceiling" models aren't documented publicly — the number shown is a conservative estimate, not a guarantee.

Transcription provider

Choose which speech-to-text service transcribes the calls. Groq (default) uses your Groq key below. My Speech API uses your own Cloudflare-hosted endpoint — Groq is still needed for scoring and analysis.

Provider

Groq API key

Used for transcription (Whisper) and every analysis call. Saved in this browser after you enter it once. Free key at console.groq.com.

Scoring (judge) model

Free-tier Groq accounts have a per-model cap on output tokens per minute. Scoring runs as 3 small calls and adapts automatically if Groq reports a cap; if a model keeps failing, pick another here.

Agri vocabulary & word correction

Speech models often mishear crop, pest, fertilizer and pesticide names. List the correct Urdu spellings here (one per line, or comma-separated; brand names in Roman are fine). They are fed to the transcriber as a hint and, if enabled, a correction pass fixes any misheard words in the transcript before speaker separation and scoring. Add your own product names, locality names and agent names for best results.

Run the word-correction pass on every transcript (1 extra analysis call per file)

Data & storage

Agents, SOPs, and call results are saved automatically to this browser's local storage — no account, no setup. This means the data lives only on this device, in this browser, and clearing your browser data will erase it. Use the backup file below to move data to another device, share it with a teammate, or keep a safety copy.

Restore from backup