WORK  /  SKELDA
CASE / 04 · 2024
Community OpsGlobal · 1k+ communities8 months · ongoing
CLIENT
Skelda

Telegram groups → managed communities.

Telegram gives you a group chat — not an operations layer. Skelda added automated protection, growth analytics, group management, and team access control. It's the platform the largest Telegram communities run on.

PROJECT META
TYPECommunity ops platform
SCOPETelegram bot + web dashboard
USERSCommunity owners + mod teams
SCALE1,000+ communities · 4M+ members
LIVE SINCEApril 2024
STATUSActive · paid SaaS
COMMUNITIES
1,000+
Protected & managed
MEMBERS COVERED
4M+
Across all tenants
AUTO ENFORCEMENT
100%
No manual triage required
ANTI-SPAM LATENCY
< 1s
From message to action
[02]The shape of it

Problem, approach, result.

I · PROBLEM

Once a Telegram community grows past a few thousand members, it stops being a chat group and starts being a small operation — moderation queues, spam waves, raids, growth tracking, role assignments, broadcasts. Telegram's native admin tools cap out around 'pin a message.' Existing bots solved one problem at a time; nobody had built the operational layer that 24/7 communities actually need.

II · APPROACH

We treated Telegram as the substrate, not the product. Skelda is a multi-tenant ops layer: each community gets a bot for in-chat enforcement plus a web dashboard for analytics, RBAC, and configuration. We built the anti-spam engine first because it's the hardest and most visible — sub-second latency, classifier-based, with fallback heuristics. RBAC was next, because once a community has a mod team, ownership of permissions is non-negotiable.

III · RESULT

Skelda now runs in 1,000+ communities covering more than 4M members. Spam is removed in under a second, with zero manual triage in steady state. Mod teams have proper RBAC instead of 'admin or nothing.' Owners can see growth, retention, and engagement in a dashboard rather than guessing. Raids that used to take down communities are auto-detected and locked down in seconds.

[03]What we shipped

The interface in production.

The live event log + RBAC view — every action the bot takes is recorded, attributable, and queryable. Mod teams operate with structured permissions, not chat-group anarchy.

skelda / @CryptoBuilders · 24,182 members
● auto-mod ON · 4 mods online
EVENT LOG · LIVE
12:04:11spam.detecteduser_8421‹wallet phishing›
12:04:11auto.removedmsg_991+ban_24h
12:04:18captcha.serveduser_8422join_flow
12:04:24captcha.passeduser_8422→ verified
12:04:51raid.detected8 accounts < 30s
12:04:52lockdown.engagedjoin_rate=0
12:05:14mod.elevateduser_5511rbac:moderator
12:06:02lockdown.liftedmanual · @rashi
12:06:18broadcast.sentmsg_101412,400 reads
12:06:32report.receiveduser_8019→ queue
LAST 24H
JOINS
412
REMOVED
67
CAPTCHAS
388
REPORTS
12
TEAM · RBAC
@rashiOWNER
@nikADMIN
@dev_samMOD
@kellyMOD
@bot_relaySERVICE

FIG. 01 / SKELDA / @CRYPTOBUILDERS — LIVE EVENT LOG + TEAM RBAC

[04]System build

What we built.

The actual surfaces, services, and capabilities that ship.

MODULE / 01
BOT

Anti-spam engine

Sub-second classifier-based enforcement. Removes phishing, raids, and noise without manual triage.

  • Classifier + heuristic stack
  • Sub-second message → action
  • Captcha gating for new joins
  • Raid detection + auto-lockdown
MODULE / 02
WEB

Owner & mod dashboard

Per-community web console for analytics, configuration, broadcasts, and team management.

  • Growth + engagement charts
  • Bot config + rule tuning
  • Broadcast scheduler
  • Searchable event log
MODULE / 03
RBAC

Team access control

Proper roles for mod teams. Audit log of every elevated action, attributable to a real account.

  • Owner / admin / mod / service roles
  • Per-permission granular control
  • Action audit log
  • API key management
MODULE / 04
DATA

Analytics layer

Growth, retention, engagement, and content trends — across one or many communities for portfolio owners.

  • Cohort retention
  • Engagement heatmaps
  • Cross-community portfolio view
  • Exportable reports
[05]Timeline

From brief to live.

Mo 101/06
Bot core + classifier v1
Built the in-chat enforcement loop with first-pass classifier on a labelled corpus.
Mo 202/06
Web dashboard alpha
Per-community setup flow + first-pass analytics + rule config.
Mo 303/06
Closed beta · 20 communities
Onboarded crypto + dev communities. Tuned classifier on real spam.
Mo 404/06
RBAC + audit log
Mod team roles, structured permissions, action audit trail.
Mo 505/06
Public launch
Self-serve onboarding, billing, support flow.
Mo 6–806/06
Scale & growth tooling
Raid detection, broadcasts, cohort analytics, portfolio dashboard.
WORDS / 06 · CLIENT
We were losing hours every day to manual moderation and the occasional raid would take the community down for an evening. Now the bot handles it before we notice. The dashboard told us things about our growth that we didn't know to ask.
Community Owner
@CRYPTOBUILDERS · 24K MEMBERS
STACK / 07
BOT
PythonaiogramRedisCelery
WEB
Next.jsReactTailwindRecharts
DATA
PostgresClickHouseETL workers
INFRA
AWS LambdaDockerK8sSentry
[]Have a similar system to build?

Tell us
what's broken.

One scoping call, one written proposal. We'll be honest if it's not a fit.

Book an architecture call