very nice idea. have some doubts so for every MCP server they defined the UI and complied into packages? so the LLM can use it. if this is the case how can it work on non defined UI's
if you want high throughput want to optimize the every component in the pipeline i try the dragonflydb pretty good comparing other database also if you add reranking like methods the ms gonna high
Let's be honest: most "AI" chatbots are just glorified search bars with a chat UI. They match keywords, pull from a FAQ, and fail the moment a user asks something slightly complex.
I learned this the hard way. A few months ago, I was a full-stack dev at an e-commerce company, and we were drowning in support tickets. We were spending 6+ hours every day answering the same repetitive questions about pricing, features, and "how-to"s. It was killing our productivity.
We looked at existing solutions, but they were either the keyword-finders that frustrated users or complex enterprise tools that cost a fortune. So, we decided to build our own.
But we quickly found that even standard RAG solutions struggle with complex questions that require understanding the customer's context. We tried hundreds of techniques, but most were only good for simple Q&A.
Our key breakthrough was developing a custom implementation on top of RAG that's designed to understand user intent and context. The goal isn't to deflect users by linking to an article, but to provide a direct, accurate answer that genuinely solves their problem. For that project, it's now accurately handling ~80% of inbound queries.
We've decided to turn this into a product and just opened the waitlist. We'd love to hear what this community thinks. Is this "glorified search bar" problem something you've experienced too?
we have FAQ section with a brochure in each category of the product but most of them want to clarify specific them about their use case (context) so they contact the support to verify this works 100% for their use case.
this is not an normal chat bot we have 100s repeated question and find a pattern within them so based on the documentation we can predict what kind of question that the user can possibly ask so based on this we have implemented our own custom RAG solution that has high accuracy with relevant context with minimal latency.
also we have human in loop concept in the upcoming update
we have thousand of product each have slight different variation for each product ex: Product EZ01 has this feature were EZO2 have a different feature in our website we have a brochure to see all the details but most of the user don't want to download and read they just open the Support and ask questions about the product and detail that was the issue we are facing
The title isn't an exaggeration. A few months ago, i was working as full-stack developer for e-commerce company on the support ticket solution part. and I were drowning in support tickets & enquires day by day.
We were spending 6+ hours every day answering the same repetitive questions about pricing, features, and basic "how-to"s. It was killing our productivity and ability to build.
We looked at existing chatbot solutions, but found they were either glorified keyword-finders that frustrated users or complex enterprise tools that cost a fortune and took weeks to implement.
So, we built our own solution. It's an AI chatbot that you can train on your own knowledge base, docs, or even just a simple FAQ page in under 5 minutes. Existing RAG solutions are not very good at complex answering we tried hundreds of different RAG techniques but most of them are good for simple questions and doesn't understand the context of the customer perspective.
The key is that it uses RAG with our our custom implementation on top to improve the accuracy and speed to find the actual answers within the customer context so users trust the responses.
For our own project, it's now accurately handling about 80% of inbound queries, freeing us up to focus on what matters.
We've just opened up the waitlist today and would love to get this community's feedback on the concept and the landing page.