Inspiration
Managing large online group chats and communities on platforms like Telegram is challenging. Toxic messages, abusive behavior, and unauthorized DM solicitation often occur faster than human admins can respond. Standard rule-based bots are often too rigid or easily bypassed, while pure LLM implementations can introduce lag.
We created Banasur - The Monitrobot to combine sub-second rule/keyword detection with agentic AI analysis, giving community managers an intelligent, ultra-fast automated moderator that handles violations instantly while keeping admins fully in control. But the features we will brag about are serverless working and very cost efficient.
What it does
Banasur - The Monitrobot acts as an active AI moderation shield for Telegram groups:
Sub-Second Auto-Muting & Removal: Detects abusive keywords or policy violations (e.g., soliciting private DMs) and automatically mutes offenders instantly to prevent further disturbance.
Rich Moderation Alerts for Admins: Instantly notifies group admins with detailed contextual alerts—including user details, flagged message context, violation reason, priority level, and AI analysis source.
One-Click Admin Actions: Provides inline interactive buttons (Unmute, Ban, Keep Muted) directly inside the admin alert feed so admins can override or confirm actions seamlessly.
Warning & Strike Tracking: Automatically maintains user warning counts (e.g., Strike 1/3) and tracks member reputation levels.
Group Health Analytics (/stats): Offers real-time metrics showing member statistics, trusted vs. flagged users, active mutes, pending reviews, and strict mode status.
How we built it
Serverless Infrastructure: Hosted entirely on Cloudflare Workers & Pages to ensure high-performance, low-latency execution across all Telegram webhook requests.
Agentic AI & LLM Integration: Built visually using Flowise AI, orchestrating Mistral AI models.
Memory & Caching Architecture: Configured custom Memory Caches and Buffer Memory nodes alongside tool agents to keep response times fast even under network stress.
Telegram Bot API: Handles real-time message monitoring, automated permissions restriction (muting), and inline keyboard management for admin alerts.
Challenges we ran into
Achieving Sub-Second Latency: Balancing thorough AI moderation analysis with instantaneous response times so toxic messages are muted before group members read them.
Handling Contextual Violations: Fine-tuning the moderation logic so the bot accurately identifies subtle off-platform solicitation (like "message me to get the link") without flagging harmless conversation.
Deployment on Edge: Ensuring seamless integration between Cloudflare edge workers, Flowise AI canvas endpoints, and Telegram webhook payloads.
Accomplishments that we're proud of
Near-Instant Action: Demonstrated real-time, sub-second content removal and automated muting during live testing.
Streamlined Admin Workflow: Built a system where group moderators can manage flags with single-click inline buttons instead of navigating complex dashboard menus.
Robust Edge Setup: Successfully deployed a serverless architecture running low-cost, scalable moderation workflows on Cloudflare.
What we learned
How to optimize serverless edge functions on Cloudflare Workers for handling high-volume webhooks.
Best practices for designing visual agent workflows in Flowise AI using memory caches and custom tool agents.
Strategies for combining hard-rule filters with LLM intent analysis to maintain speed without sacrificing accuracy.
What's next for Banasur - The Monitrobot
Multi-Platform Expansion: Adapting Banasur's moderation agent for Discord servers and Matrix communities.
Multimodal Moderation: Incorporating computer vision models to flag toxic images, NSFW media, and graphical spam.
Custom Admin Dashboard: Building a dedicated web UI for community owners to customize strictness thresholds, specific blocklists, and strike rules.
Built With
- cloudflare
- flowise
- typescript
Log in or sign up for Devpost to join the conversation.