Bulkbeat TV - AI-Filtered Telegram Market Intelligence Bot
A production Telegram bot built for a Sitekraft.dev client engagement that monitors NSE filings, market news, and bulk deals, enriches documents with OCR, scores events with AI, and sends high-signal alerts.
Overview
Bulkbeat TV is a production-grade Telegram market intelligence system built during a Sitekraft.dev client engagement. The goal was not to create another noisy alert bot, but to build a disciplined pipeline that could monitor market-moving information, reduce false urgency, and deliver only the updates that deserved a trader's attention.
I owned the implementation end to end, covering ingestion logic, AI filtering, alert formatting, admin controls, subscription workflows, and deployment-oriented reliability decisions.
The Problem
Serious market participants do not struggle with lack of information. They struggle with overload. NSE filings, SME disclosures, bulk deals, and financial media updates can arrive in rapid bursts, and most of them do not deserve an interruption.
The product challenge was to turn that stream into a selective intelligence feed:
- watch multiple sources continuously
- extract useful context from raw filings
- suppress duplicate or low-signal events
- deliver output fast enough to be actionable inside Telegram
What I Built
The final system combines five layers into one product workflow:
- Multi-source monitoring across NSE announcements, NSE SME, bulk deals, and supporting market sources
- Document enrichment through PDF parsing and OCR fallback for image-based filings
- AI-based scoring to rank how relevant or market-moving an event appears
- Strict alert controls using filtering rules, cooldowns, and caps
- Telegram delivery for both subscribers and admin operations
This design keeps the product focused on signal quality instead of alert volume.
Core Features
- Real-time monitoring of NSE and related market sources
- Async ingestion pipeline for frequent polling without blocking the bot
- PDF extraction with OCR fallback for hard-to-read documents
- AI-assisted impact scoring before an alert is allowed through
- Structured Telegram messages optimized for fast reading
- Separate admin workflows for live monitoring, grants, and broadcast actions
- Subscription handling with payment verification and market-day access logic
- Operational safeguards such as backups, health checks, and single-instance protection
Technical Approach
- Runtime: Fully async Python architecture for concurrent polling, enrichment, and delivery without blocking
- Data collection: Multi-source async HTTP layer with HTML parsing for structured and unstructured market data
- Persistence: Lightweight embedded database configured for safe concurrent writes on constrained infrastructure
- Scheduling: Event-driven scheduling layer for ingestion cycles, billing checks, and daily maintenance windows
- Document processing: PDF text extraction with image-based OCR fallback for complete filing coverage
- AI layer: Large language model integration for event impact scoring and relevance classification
- Operations: Process health monitoring, automated backups, and single-instance safeguards for 24/7 stability
Why This Project Stands Out
What makes Bulkbeat TV more interesting than a typical automation bot is the amount of judgment built into the pipeline. The hard part was not fetching information. The hard part was deciding what not to send.
That meant designing:
- aggressive noise suppression
- source-aware filtering
- low-resource runtime behavior
- admin tooling for real operational control
- a premium-feeling subscriber experience inside Telegram
Challenges and Tradeoffs
1. Signal quality versus speed
Fast alerts are useful only if they stay trustworthy. I solved this by placing AI scoring and rule-based gating in front of delivery, so speed did not come at the cost of alert quality.
2. Messy filing formats
Many filings are inconsistent or difficult to parse directly. To avoid losing context, I added OCR-backed enrichment when direct extraction was not enough.
3. Reliable 24/7 runtime on constrained infrastructure
The bot was designed to remain practical on a small VPS. Async I/O, SQLite WAL, cooldown logic, health checks, and maintenance jobs helped keep the system stable without bloating infrastructure cost.
4. Operational control
A production bot needs more than a user-facing interface. I implemented admin-side controls for monitoring, access management, and broadcast workflows so support and operations would not depend on manual intervention.
My Ownership
- Project Manager at Sitekraft.dev — owned the full delivery end to end
- Defined the alerting workflow, system behavior, and product logic
- Built the backend automation and Telegram experience
- Implemented billing and subscriber access flows
- Designed for long-running deployment and operational safety
Public-Safe Positioning
Some implementation details remain intentionally generalized in public because the real value of the product is in its signal policy and operational tuning. This case study highlights the engineering depth and system design without disclosing the parts that make the workflow easy to replicate.
Outcome
Bulkbeat TV became a deployable Telegram intelligence engine built for continuous monitoring, controlled delivery, and premium subscriber usage. It demonstrates product ownership across automation, AI integration, document processing, payments, and production reliability in one cohesive system.