Ask Alvi Advanced search engine || Clone of Perplexity|| Discord Ai bot || Telegram Ai bot
Ever wondered what happens when you ask a question on our AI-powered platform? Here’s how it all works:
Step 1: First, We Clean Up Your Question
When you type in a question, the AI doesn’t just take it as is. First, we polish it up using Groq’s Mistral model and OpenAI’s completion model. This makes sure your question is clear and well-structured so that we get the best possible results in the next steps.
Step 2: Hunting for Answers on the Web
Next, we search the internet—but not just with Google. We use:
🔹 Google (via Serper API)
🔹 Brave Search (for more independent results)
🔹 Qwen-2 LLM (which also gives AI-powered insights)
These sources help us find the most relevant articles, research papers, and discussions related to your question. Instead of just giving you links, we dig deeper to extract useful information.
Step 3: Scraping & Cleaning Up the Data
Once we have the top links, we don’t just rely on summaries. Instead, we scrape the full webpage using Bright Data’s web scraper.
But raw web pages are messy—so we clean up the text using Cheerio, which removes extra code, and all the unnecessary clutter.
Step 4: Breaking Down the Information
Now, we have a bunch of clean text, but it’s still too long to process in one go. So, we split it into smaller chunks using Recursive Character Text Splitter. This helps the AI focus on the important details without losing context.
Step 5: Storing the Knowledge for Future Use
Each chunk of text is then converted into smart embeddings using OpenAI’s embedding model and saved in a memory vector database.
This allows the AI to remember past information and pull up relevant details instantly whenever needed.
Step 6: Finding the Best Matches
Before answering, the AI searches the memory database to find similar chunks of information. This ensures that the final response is based on accurate and relevant knowledge, instead of just random guesses.
Step 7: Generating the Perfect Answer
Now comes the fun part! The memory vectors response is sent to the AI models like:
🔹 Groq (Mistral) – Fast and efficient
🔹 Llama – Great for deeper reasoning
🔹 Claude 3 – Best for well-structured responses
With all the pieces in place, the AI puts everything together and generates a final response that’s accurate, relevant, and well-explained and
Step 8: Follow up question
Now based on the answer the asnwers are sent to the ai model for follow-up questions.
And Boom! There’s Your Answer!
All of this happens in just a few seconds—from refining your question to searching the web, scraping data, analyzing information, and generating a response. Pretty cool, right? 🚀
Links -
1. Website - https://ask-alvi.vercel.app/
2. Discord bot - https://discord.com/oauth2/authorize?...
3. Telegram Bot - https://t.me/ask_alvi_bot
Github Links -
1. Website - https://github.com/alvi00/ask_alvi
2. Disocrd bot - https://github.com/alvi00/ask_alvi_di...
3. Telegram bot - https://github.com/alvi00/ask_alvi_tel
#AI #Tech #MachineLearning #RAG #OpenAI #AIworkflow #ArtificialIntelligence #TechInnovation #Groq #Claude3 #LlamaAI #askalvi #Askalvi #ASKALVI