A fintech founder needed an AI chatbot that gives accurate financial answers from their own knowledge base, not generic LLM responses.
We built an embeddable, RAG-powered conversational AI for domain-specific financial guidance. The system retrieves answers from the client's own knowledge base, ensuring accuracy over generic LLM hallucination, while keeping running costs under $10/month.
Client
A fintech startup building...
Timeline
9 weeks
Team
2 engineers
Industry
Fintech
The client ran a financial services platform and needed visitors to get accurate, domain-specific answers to finance questions. Generic chatbots either gave wrong information or hallucinated financial data, which is a liability in this space. They needed something that pulls from their own vetted content — PDFs, articles, FAQs — and presents it conversationally. It also had to embed seamlessly into their existing site without breaking styles, and the whole thing needed to run affordably on a single VPS.
Built a Go backend with clean architecture, PostgreSQL + pgvector for storing both relational data and vector embeddings in a single database
Implemented a RAG engine that processes uploaded documents (PDF, DOCX), splits them into semantic chunks, generates embeddings, and retrieves relevant context for each query
Integrated Gemini Flash Lite for cost-efficient LLM responses with streaming via Server-Sent Events, plus finance-specific system prompts and auto-appended disclaimers
Created an embeddable Preact + Shadow DOM widget (<30KB) that drops into any site via a single script tag, with full style isolation
Built a React admin dashboard for managing the knowledge base, viewing conversations, monitoring usage, and configuring system prompts
Added domain-specific financial calculators — EMI, SIP, compound interest, retirement projector — with an intent router that detects when users need calculations vs. information retrieval
<30KB
Widget Bundle Size
~$10/mo
Running Cost
9 weeks
Delivery Timeline
<2s
Widget Load Time
Widget loads in under 2 seconds and works on any site without style conflicts
RAG accuracy significantly higher than generic chatbot responses for domain-specific queries
Monthly running cost stays under $10 on a Contabo VPS
Admin team can update the knowledge base without developer involvement
“They understood the problem immediately — we couldn't have a chatbot giving wrong financial info. The RAG approach means our answers are always grounded in our own content, and the whole thing costs less than a Netflix subscription to run.”
— Founder, Fintech Platform
We've solved problems like this before. Let's talk about yours.
Start a ConversationTell us about your project. We'll give you honest feedback on scope, timeline, and whether we're the right fit.
Start a Conversation