Sathi (साथी) — Building an AI Companion for My Sister's CA Journey ⭐

yuvaraj — Mon, 05 Oct 2026 06:50:14 +0000

Sathi (साथी) — An AI Companion for My Sister's CA Journey

This is my submission for the Hacktoberfest Weekend Challenge: Build for a Friend.

The person I built this for

My sister is preparing for Chartered Accountancy.

I noticed that the difficult part wasn't only learning accounting, taxation, auditing, or law.

There was a much bigger problem:

There was no single companion helping her navigate the entire journey.

She had books, study material, chapters, revision, tests and deadlines — but no system that could explain a difficult concept, remember the conversation, track her progress, help her organize study tasks, and give her a small mental break when preparation became overwhelming.

So I built Sathi (साथी).

Sathi means companion.

What is Sathi?

Sathi is a personal AI CA companion designed around four things:

🤖 AI-powered CA study companion
📚 Syllabus and chapter tracking
📊 Progress and daily study planning
🌿 Sanctuary for breathing and mental grounding

The goal isn't to replace teachers or professional CA advice.

The goal is to give a student a companion that is available whenever they sit down to study.

The AI

Sathi uses Google Gemma 2, an open-weight model, running locally through Ollama.

The model runs on my laptop's NVIDIA RTX GPU rather than sending every study question to a closed AI service.

The architecture is:

React + Vite

↓

FastAPI

↓

AI Provider abstraction

↓

Ollama

↓

Gemma 2

↓

NVIDIA GPU

This also means the core study conversation can run locally without sending the student's questions to a third-party AI API.

Why open-weight AI mattered for this project

This was one of the most important architectural decisions.

CA preparation can involve personal study questions, frustrations, mistakes, and doubts.

Using an open-weight model allowed me to experiment with a local AI companion where:

inference can happen locally
there are no per-question model API costs
the model can be replaced or adapted later
the AI layer isn't locked to a single closed provider
the application can continue evolving toward a privacy-first study companion

For me, open AI wasn't just a technology choice.

It made the idea of a truly personal companion possible.

The Study Companion

A student can ask Sathi something like:

"Explain depreciation in simple terms."

Then continue:

"Give me an exam-oriented example."

The conversation maintains context, so the student doesn't have to repeatedly explain what they were discussing.

Sathi is also designed to explain concepts in a student-friendly way instead of simply returning textbook-style answers.

Study & Progress

The application includes:

subject and chapter tracking
chapter status
revision notes
daily study tasks
study streaks
progress visualization
direct connections between chapters and the AI companion

The idea is to turn a huge CA syllabus into smaller, actionable pieces.

Sanctuary

CA preparation isn't only an academic challenge.

There are long study sessions, exam pressure and moments where students feel like they're falling behind.

Sathi therefore includes a Sanctuary with:

guided box breathing
grounding experiences
negative-thought reframing through Gemma

It's intentionally not designed as a therapist or medical system.

It's simply a small reminder that the person studying for the exam is more than their exam result.

Technology

Frontend

React
Vite
CSS
React Markdown

Backend

Python
FastAPI
SQLite

AI

Google Gemma 2
Ollama
Local GPU inference

Voice

ElevenLabs integration

Architecture

I also used an AI provider abstraction so the application isn't tightly coupled to one inference provider.

This allows the local Gemma implementation to evolve toward other providers or deployment environments later.

What I learned

The biggest lesson from building Sathi wasn't actually about adding another AI chatbot.

It was about designing AI around a real person and a real problem.

Starting with my sister changed the design decisions.

Instead of asking:

"What features can I add to an AI app?"

I started asking:

"What would actually help her tomorrow when she sits down to study?"

That led to the combination of AI tutoring, progress tracking, study planning and a space to step away from the exam.

Demo

🎥 Demo Video:

[https://drive.google.com/file/d/1mxjUOozd5g4_Vtl4AKpH3hi0AjxFqSWc/view?usp=sharing]

💻 Source Code:

https://github.com/YUVA56-gomu/Dev-Challenge-01

What's next?

Sathi is currently an MVP.

The next evolution would be deeper CA knowledge grounding through RAG, a complete ICAI syllabus knowledge base, richer student memory, improved voice interaction, and eventually personalization based on the student's learning patterns.

But the most important part is already here:

I built something for someone I care about.

Sathi is my attempt to make the CA journey feel a little less lonely.

devchallenge #weekendchallenge #hf26challenge

excited joining hacktoberfest 2026😊😙

yuvaraj — Thu, 01 Oct 2026 17:05:11 +0000

DEV Community: yuvaraj