我为朋友的祖母 Malti Devi 打造语音助手 Saarthi,用 Gemini API 让她打电话提问
I built Saarthi, a voice assistant that phones my friend Krishna's grandmother, Malti Devi and for SBI customer's
开发者用 Java + Spring Boot 构建语音助手 Saarthi,通过 Gemini API 将语音转文字、理解问题并生成语音回复,让不使用触屏手机、独居的 Malti Devi 只需说话就能获得答案,全程无需打字、屏幕或菜单。界面同时支持文字与语音两种模式。
This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend
What I Built
I built Saarthi for my friend Krishna Singh's grandmother, Malti Devi and SBI cutomer's . Talk about only Malti Devi here she lives alone, and she doesn't use a touchscreen phone. Most apps today assume you can tap, type and swipe, so for someone like her, solving even a small problem online is almost impossible.
Saarthi fixes that with just her voice. The AI agent actually calls her. If she wants to talk or ask something, she just speaks, and the agent understands her, finds the answer and talks back. There is no typing, no screen, no menus.
Demo
Code
PARAKH — परख
parakh (परख): "to assess, to gauge the true quality of something."
A transparent, deterministic, network-free candidate ranker for the Redrob × Hack2Skill Intelligent Candidate Discovery & Ranking challenge. It ranks the top 100 candidates out of 100,000 for the Senior AI Engineer — Founding Team role using an explainable panel of feature-engineered scorers — no embeddings, no ML model weights, and zero network calls during ranking.
Every score is fully decomposable and every line of the reasoning column is
assembled only from real fields on the candidate's profile, so hallucination
is impossible by construction.
TL;DR — reproduce the submission
Single command (local, requires JDK 17):
./gradlew bootJar && java -jar build/libs/parakh.jar <path-to>/candidates.jsonl ./submission.csv
Single command (Docker — no JDK needed, runs with the network physically off):
# 1. put the contest file at ./data/candidates.jsonl mkdir -p data && cp <path-to>/candidates.jsonl ./data/
…
How I Built It
The backend is written in Java with Spring Boot. I used the Gemini API for the AI part: it converts her speech to text, works out what she is asking, and generates the reply. The reply then goes back to her as speech, so she hears the answer instead of reading it.
The interface works in both modes, text and speak, so anyone can use it either way.
Why Does Open Innovation Matter?
Most technology is built for people who can use a touchscreen. People like Malti Devi are left out, and they are often the ones who are alone and need help the most. Saarthi shows that a simple voice call can give them access to the same answers everyone else gets.
What Krishna Said
When I told Krishna about it, he was shocked. He was really happy, and it was a very emotional and enjoyable moment for him.
Thanks for giving me this opportunity
来源:Google AI:DEV 作者专属(RSS) · dev.to