I’m excited to share Docu Talk, a Streamlit-powered web app that lets you create custom Chat Bots from your PDF documents in just seconds!
Key Features: AI-Powered Responses: Structured and formatted replies based on document content Cloud Integration: Secure communication with cloud-stored files via Gemini Machine Learning Optimization: Predicts response times using historical data
Streamlit-Powered Frontend: Google/Microsoft/Email Authentication (thanks to Streamlit 1.42.0!) Session & Cookie Management for user persistence Dynamic HTML Components for an interactive UI
Security & Deployment: Encrypted Password Storage with hashing Whitelisted Database Access via VPC/NAT for enhanced security Cloud Deployment on Google Cloud Run & Streamlit Cloud Automated Email Service with AWS SES Secure File Storage using Google Cloud Storage
Code Quality & Best Practices: Linted & Verified with ruff Modular Architecture (authentication, pages, services…)
Thank you @mojsilo ! Yes, there is no RAG, but “in-context processing”, capable of handling up to 1,000 pages of documents with Gemini!
In-context processing is becoming increasingly interesting compared to RAG for several reasons: models can accept longer inputs, generate faster, and are becoming less and less expensive
Hey @Antoine3 , really interesting work! I recently came across Davia (Introduction - Davia Starter Kit), which simplifies UI creation for chatbots. Do you think it could have been useful for your project?
Thank you for your message, @Fuad! Here is the detailed breakdown of the billing:
To give you some context, I initially deployed the application on Cloud Run for a few weeks. Later, I moved it to Streamlit Cloud in hopes of gaining some visibility through the Streamlit Gallery or by convincing Streamlit to communicate about the app.
If I had stayed on Cloud Run, my monthly costs would have been around $10–$15. Currently, on Streamlit Cloud, my total monthly expense is $5~$10.
Hosting:
While using Cloud Run, I configured an instance with 2 CPUs & 4GB RAM, setting min_instances=0, meaning the instance shuts down when there are no active users. For 20~30 monthly users, this cost around $3/month.
However, if the instance were always running (min_instances=1), the cost would be $100/month for the same configuration.
VPC (Virtual Private Cloud):
To ensure secure communication between Cloud Run and MongoDB Atlas, a static IP is required, which is configured via Cloud NAT. This results in a $3/month networking fee.
AI Model Costs:
I use Gemini from Vertex AI for the generative AI functionality. With 20~30 monthly users, this amounts to approximately $5/month.
Database:
I use a free-tier MongoDB instance with storage under 500MB, so there’s no additional cost.
Email Service:
I purchased a domain name for $3 and use AWS SES for email services. AWS SES is free up to a high number of requests, which is more than sufficient for my needs.
As a solo developer, I’ve spent some time optimizing the infrastructure to keep costs as low as possible in the long run. I’m confident that I can keep the application running smoothly for the future.
That’s interesting, I wasn’t familiar with this tool. I chose to build the application in Streamlit because I have advanced expertise in it and knew I could develop it very quickly (in less than 15 days).
Additionally, I needed custom features beyond just the chatbot itself, such as a login page, managing chatbot settings and a chatbot creation form.
Key Features:
AI-Powered Responses: Structured and formatted replies based on document content
Try the App: https://docu-talk-ai-apps.streamlit.app
Explore the Open-Source Code: GitHub Repo
