🎙️ Voice-Enabled Task Tracker

A full-stack task manager with voice input, AI-powered task parsing, and a clean React UI.

---

🚀 1. Project Setup

Prerequisites

• Node.js ≥ 18
• MongoDB (Atlas preferred)
• Ollama installed with model gemma2:2b or relevant models like gemma3:1b
• npm or bun

---

Installation

Clone Repository

git clone <repo-url>
cd <project-folder>

---

Backend Setup

cd backend
npm install

Create backend .env:

PORT=3000
MONGODB_URI=<your-mongodb-uri>
OLLAMA_HOST=http://127.0.0.1:11434

Start backend:

npm run dev

---

Frontend Setup

cd frontend
cd vite-project
bun install

Start frontend:

bun run dev

---

Running Everything Locally

Service	Command
MongoDB	Run Atlas
Ollama	ollama serve
Backend	npm run dev
Frontend	bun run dev

---

🛠️ 2. Tech Stack

Frontend

• React + Vite using TS
• TailwindCSS
• Web Speech API

Backend

• Node.js + Express
• MongoDB
• Ollama (Gemma2:2b) for task parsing

AI

• Local LLM (Gemma2:2b)
• Multi-tier parsing (AI → regex → keywords)

---

🧩 3. Decisions & Assumptions

Design Decisions

• 3-tier parsing: LLM → regex → keyword fallback
• Gemma2:2b model for fast local inference
• Frontend uses Web Speech API for browser-native voice capture

Key Assumptions

• Natural English speech input
• Users may not specify priority/due date → sensible defaults
• Dates stored in UTC

---

🤖 4. AI Tools Usage

Tools

• GitHub Copilot – boilerplate & autocomplete in VS Code
• Claude – architecture suggestions, parsing logic

How They Helped

• Suggestions in multi-stage task parsing strategy
• Helped craft prompts for consistent JSON output from Ollama
• Guided date-parsing logic & fallback handling

What I Learned

• Ensuring deterministic AI outputs requires explicit instructions
• Always add non-AI fallback paths
• Smaller local LLMs are sufficient for structured extraction tasks

---