Powered by Llama 3.3 70B

My Very First End-to-end AI Project: Codexa.

I set out to build an efficient, budget-friendly AI app, but the journey taught me a crucial lesson: balancing performance and cost is rarely simpleβ€”it requires navigating deep technical trade-offs.

6+
Languages Supported
RAG
Powered Analysis
fibonacci.py
function fibonacci(n) {
  if (n <= 1) return n;
  return fibonacci(n-1) + fibonacci(n-2);
}
Codexa
AI Response

Recursive Function: This is a recursive implementation of the Fibonacci sequence. It returns the nth Fibonacci number by recursively calling itself...

recursionalgorithmO(2^n)

Understand Code Faster

Built for developers, students, and anyone who wants to understand code faster.

Codexa Preview 1
Codexa Preview 2
Codexa Preview 3
Codexa Preview 1
Codexa Preview 2
Codexa Preview 3
96%

Latency Reduction

Reduced response time from 15s to <500ms using real-time LLM token streaming via SSE.

15s β†’ <500ms
95%+

Relevance Accuracy

Production RAG pipeline with LangChain and Llama-3.3-70B achieving high precision at 75% similarity threshold.

pgvector powered
<50ms

Embedding Speed

Local embedding generation using Xenova Transformers.js, eliminating external API costs entirely.

Zero API costs
42%

Query Optimization

Optimized with embedding model preloading and lightweight 384-dim vector embeddings.

384-dim vectors
Powered by Llama-3.3-70B
Real-time SSE Streaming
LangChain RAG Pipeline
pgvector Similarity Search
Live Demo

See Codexa in Action

Watch how Codexa transforms complex code into clear, understandable explanations with just a few clicks.

πŸ”’codexa-ai.site
Open App
Codexa Demo

Raihan Rizki - Personal Portfolio Website