🦇

Alfred

AI-Powered Knowledge Butler

Capture everything. Forget nothing. Let Alfred handle the rest.

React Native Expo SDK 54 Appwrite Gemini 2.5 Qdrant Vector DB Spaced Repetition

3

Input Modalities

8

Cloud Functions

3072-dim

Vector Embeddings

github.com/sidx007/alfred

Overview

What does Alfred do?

Alfred is a personal knowledge management app that lets you dump raw information — text, audio, or images — and automatically turns it into a structured, searchable, and reviewable knowledge base. Think of it as a second brain with a butler personality.

📝

Capture

Text, audio recordings, and images — all processed automatically on upload.

🧠

Process

Transcription, OCR, semantic chunking, topic classification, web enrichment, vector embedding.

🔁

Review

Daily spaced-repetition reports from chunks captured 1, 3, 5, and 7 days ago — written in Alfred's voice.

🃏

Flashcards

Anki-style cards with Again / Good / Easy grading, auto-generated from your notes.

💬

RAG Chat

Ask questions — semantic search retrieves relevant memories for context-aware answers.

🎙️

Voice

Live bidirectional voice chat via Gemini 2.0 Flash WebSocket streaming.

Architecture

How it's built

📱 Mobile App
React Native / Expo

🌐 Web Dashboard
Vanilla JS + Express + Vite

↓

Appwrite Cloud Functions (Python)

audioFunction imageFunction clusteringFunction processSegmentFunction vectorEmbedFunction reportGeneratorFunction dailyReportPipeline vectorRetrieveFunction

↓

Qdrant Cloud — Vector DB (3072-dim)
memory · knowledge_base · topics · flashcards · daily reports

LLM Routing

Right model for each task

Alfred doesn't use one model for everything — each step in the pipeline uses the model best suited for that specific task.

Task	Model
Speech-to-Text	Deepgram Nova-3
Image OCR	OCR.space
Semantic Clustering	Groq / Llama 3.3 70B
Topic Classification	Groq / Llama 3.1 8B
Report Generation	Gemini 2.5 Flash
Web Research	Gemini 2.5 Flash + Google Search
Live Voice Streaming	Gemini 2.0 Flash (WebSocket)
Vector Embeddings	Gemini Embedding 001 (3072-dim)

Pipeline

From capture to knowledge

1

CaptureText typed, audio recorded, or image taken on mobile app

2

Transcribe / OCRDeepgram for audio, OCR.space for images → raw text

3

Semantic Chunk + ClassifyLlama 3.3 70B splits text → Llama 3.1 8B assigns topics

4

Enrich + EmbedGemini web-researches each chunk → Gemini Embedding 001 → stored in Qdrant

5

Daily Reports + RAG Chat + FlashcardsSpaced repetition, semantic Q&A, and Anki-style review — all over the same Qdrant store

"Shall I remind you, sir, that you studied distributed systems three days ago and appear to have retained absolutely none of it? Allow me to refresh your memory."