Private game ecosystem · A whole business one operator can run.
↓ SCROLL FOR DETAIL
A private ecosystem for a game
Telegram-native operations layer for a live game project running across three branches. One operator runs analytics, traffic, donations, moderation, and partnerships through a single control plane instead of juggling scattered chats and half-dashboards.
The ecosystem is not a collection of unrelated bots. It is a connected control layer built around two constant product jobs: analytics keeps the project understandable, traffic keeps it alive. Everything else exists to support one of those jobs.
Before it was formalized, metrics lived across chats and partial dashboards. Traffic passed through unstable surfaces. Donations were split across multiple gateways. Volunteer and moderation mechanics depended on manual coordination. The system compresses those moving parts into something one person can actually operate.
Architecture
The mono-repo as a navigable tree. Click a contour to see its role, stack, and scale. Expand it to see the services inside. Branch runtimes (classic, lite) are separate game versions with their own audiences, shown as full contours rather than folded into main.
ecosystem/131,378 LOC total
core_agent/56,006 LOC
main/39,362 LOC
apps/portal/23,490 LOC
apps/donate_bot/7,794 LOC
apps/aux_donate_bot/3,589 LOC
apps/traffic_bot/2,935 LOC
administration/24,059 LOC
apps/support_bot/3,113 LOC
apps/ban_bot/1,119 LOC
apps/chat_moderation_bot/3,712 LOC
apps/partnership_bot/13,582 LOC
apps/anti_cheat_api/653 LOC
branch_legacy/9,188 LOC
lite/1,611 LOC
infrastructure/1,152 LOC
docs/5 docs
core_agent/
Central AI operations agent
Reads live project state, answers natural-language admin questions, generates charts and reports, keeps short-term and persistent follow-up context. Audits three chat categories continuously and derives hate / hype signals. LLM as decision surface, deterministic tools as execution.
The portfolio value is not that there are many bots, it is that bots, portal, and control logic were turned into one ecosystem with clear responsibilities. This demonstrates systems thinking, operational design under constraint, product architecture for live traffic and analytics, and the ability to keep a complex project operable as a single person.
A few runtime stories
01
After an attack the database, 7M player records, was wiped. Restored the live state solo in one week, no downtime visible to users, moderation and donations kept running off the hot paths while the cold state was rebuilt.
02
One admin operator generates up to 200 analytical summaries per month through natural-language questions instead of dashboard switching. The agent keeps follow-up context across those summaries.
03
Donations flow through multiple gateways with unified state tracking. 0.1M lifetime paying users, $2 average check, $170k handled end-to-end.
04
Moderation processes ~5,000 messages per day across the main chat. Cheater reports are reviewed by volunteers through a structured runtime, 30k cases handled in three months, surfaced back to the owner as an operational signal, not a message queue.
Also in these roles
System builder is the primary lens. The day-to-day cuts across the other roles too.
AI engineer
Central operations agent on Gemini + Mongo: 80 admin operators issue natural-language requests, agent runs deterministic tools and SQL under the hood, keeps short-term plus persistent follow-up context across reports, audits three chat categories continuously and derives hate / hype indices. Retries and fail-closed semantics are explicit, no silent fallback.
Founder
Solo ownership of a live product: live money through a unified multi-gateway donation contour ($170k handled, $2 avg check, 0.1M lifetime paying), live moderation (5k messages a day, 30k cases handled in 3 months, cheater reports included), live partnerships (4k creator videos, 2k volunteer applications). Every architectural call had to survive real users, real incidents, real revenue.
Fullstack
One person across backend (Python + Node.js, Mongo + MySQL + Postgres), schema, the unified donation contour, admin UI, analytics surfaces, and the public web shell on Next.js + TypeScript. Mono-repo, 131k LOC.
Source code is under NDA. Project name is withheld. What is shown here is the information architecture, safe scale signals, and the operational shape, not the sensitive internals.
On a serious offer I can walk through specific parts of the system privately, e.g. the agent, donation contour, or moderation runtime, live and unfiltered, under a discussion-level NDA.
