Skip to content

Sentient

**KhataAgent** is a local-first AI ledger assistant for kirana stores that works even without internet.

Core trick

Gemma 4 runs on-device via LiteRT-LM to parse spoken transactions into a validated SQLite ledger, asking the user to confirm uncertain entries.

Tags

Description

**KhataAgent** is a local-first AI ledger assistant for kirana stores that works even without internet. Built as a voice-driven agent on **Gemma 4 E2B** running entirely on-device via LiteRT-LM, it understands Hindi, Kannada, and English, converts spoken transactions into validated ledger entries, stores them in SQLite, and safely recovers from errors by asking for human confirmation when uncertain. When connectivity is available, an optional cloud agent performs heavier tasks such as weekly reconciliation, anomaly detection, and inventory insights, keeping day-to-day bookkeeping fast, private, and reliable even in offline environments. Talk to your AI for complex report or questions when you're connected to network.
shani-singh1/ledger-deepmind-hackathon★ 10 forksKotlinCreated Jul 11, 2026 · last push Jul 11, 2026

README excerpt

# 🧾 KhataAgent — the ledger agent that works where the internet doesn't

**An offline, voice-driven ledger + inventory agent for India's 13-million kirana (corner) stores — running Gemma 4 E2B fully on-device via LiteRT-LM.** Speak a transaction in Hindi/Kannada/English, and the agent parses it, validates it in code, writes it to SQLite, recovers from its own mistakes, and **defers to the shopkeeper when it's unsure** — all with the phone in airplane mode.

> **One-line pitch:** an agent that works where the internet doesn't, and knows exactly when it needs help.

### 🎥 60-second demo → **https://youtube.com/shorts/kKpmK_VZs9I**

**Tracks:** 🏆 Special Prize — *Best Use of Gemma 4 (Local-First Agents)* · 🤝 *Problem Statement 2 — Autonomous Orchestration* (via the cloud-escalation + live-AI layer).

---

## ⚡ Why it wins in 20 seconds

| | |
|---|---|
| 🔒 **100% offline core** | Gemma 4 E2B runs **on the phone's GPU** (OpenCL, ~50 tok/s). Airplane mode → still works. No cloud dependency for the ledger. |
| 🗣️ **Voice-first, 9 Indian languages** | Speak or type in Hindi, Tamil, Telugu, Kannada, Bengali, Marathi, Gujarati, Malayalam, Punjabi or English. The whole app localises. |
| 🧠 **Real agent loop** | sense → decide → act → **check** → defer. The "check" is deterministic Kotlin, not the model. Every uncertain call becomes a human-confirm card. |
| 🌐 **Knows when it needs help** | Online, it automatically escalates to **Gemini** — including a **live voice chat** that answers c

Team