# Building FinSaathi: A Voice-First AI Financial Assistant with LiveKit and Murf

> Source: <https://dev.to/truebrace05/building-finsaathi-a-voice-first-ai-financial-assistant-with-livekit-and-murf-4pfe>
> Published: 2026-08-15 18:29:43+00:00

Financial information can be difficult to understand. Banking terms, loans, credit scores, payments, and other financial decisions can quickly become overwhelming when users have to navigate everything through forms and complicated interfaces.

So I wanted to explore a simpler interaction:

What if financial guidance could start with a conversation?

That idea became **FinSaathi**, a voice-first AI financial assistant.

The first goal was simple: get a real-time voice assistant working end-to-end and deploy it.

The current architecture is:

**Next.js Frontend → LiveKit → Python AI Agent → Voice/AI Services**

The frontend is deployed on Vercel, while the LiveKit agent is deployed on Railway.

Users can open the application, start a conversation, and interact with the FinSaathi agent through voice.

Getting the agent to work locally was relatively straightforward.

Getting the same system to actually run in production was a different problem.

The Railway deployment initially failed with:

python: can't open file '//src/agent.py':

[Errno 2] No such file or directory

The problem turned out to be related to how the application path and startup command were being handled inside the Docker deployment.

After fixing the container and Railway startup configuration, the deployment moved further — and exposed another issue.

Because the container runs the application as a non-root user, UV initially could not create its cache directory:

Permission denied: '/app/.cache/uv'

Fixing the permissions allowed the actual **LiveKit AgentServer** to start successfully.

The production logs then showed the agent listening for connections and registering its worker with LiveKit.

That was the first real milestone: the backend was no longer just **"working on my machine"** — it was actually running in production.

After the backend was live, I tested the frontend against the production LiveKit setup.

The voice interaction worked end-to-end.

The user can speak to FinSaathi, the session connects through LiveKit, and the deployed agent processes the conversation and responds through the voice pipeline.

This was the main goal of the first phase.

FinSaathi also has a SQLite-based memory and call analytics layer.

The local application works with this storage layer, but deployment exposed a file-permission issue when the application attempted to write to the SQLite database during session completion.

So this part is **not being treated as completely solved yet**.

The next step is to move this production storage layer to a more appropriate persistent database/storage setup instead of relying on a writable SQLite file inside the application container.

The current interface is intentionally focused on the core voice experience.

The next phase will be about turning the working prototype into a more complete product.

Planned improvements include:

The escalation and analytics pieces are being designed around the idea that an AI assistant should know when a conversation needs human support rather than trying to handle every situation itself.

The core voice system is now deployed.

Currently working:

Still being improved:

[https://finsaathi-alpha.vercel.app](https://finsaathi-alpha.vercel.app)

[https://github.com/true-brace05/murf-livekit-starter](https://github.com/true-brace05/murf-livekit-starter)

This is only the first phase of FinSaathi, but getting the complete voice pipeline from a local prototype to a deployed system was a valuable engineering milestone.

The next phase is about building the product around that core.