PHANG builds systems that eliminate operational blindness, trust failures, and fragmented execution in markets where those failures have the highest cost. Each product is an infrastructure layer — not an interface. The moat is what the system knows, enforces, and remembers.
Document intelligence and analysis pipeline for institutional workflows. Replaces fragile manual interpretation with structured reasoning over unstructured data — CSV analysis, document intelligence, AI-assisted writing. Designed for operators who process high volumes of documents under time and accuracy constraints.
Trust and coordination infrastructure for fragmented African housing markets. Enforces full cost disclosure, agent identity verification, and complaint escalation at the system level — not as policy, but as architecture. WhatsApp-first search for markets where web friction is a real barrier.
Adaptive strategic intelligence for long-term human execution systems. Models consequences, detects failure cascades before they compound, and rebuilds execution plans when environmental conditions change. Not a task manager. An execution environment with memory and consequence awareness.
Operational command interface for the PHANG infrastructure layer. Two-layer architecture: deterministic slash commands for known operations, DeepSeek reasoning for diagnostic queries. Docker socket access, event telemetry, container lifecycle control — all through Telegram.
One VPS. One PostgreSQL instance shared across products. One gateway routing all traffic. No managed services that abstract operational visibility away from the operator. The system is owned end-to-end.
All products emit to a shared ops_events table. System state is observable from a single query surface. The CUA reads this table to reason about infrastructure health without external monitoring dependencies.
No OpenAI. No Anthropic dependency for production inference. DeepSeek is called directly from product code with deterministic context injection. The AI layer reasons over real operational data, not hypothetical inputs.
New infrastructure is only introduced after a seven-question reuse audit confirms no existing component can be extended. Complexity is not added to signal maturity. Four percent RAM utilization under real load is the target, not the floor.
In markets where smartphone penetration outpaces web adoption, WhatsApp is not a notification channel — it is the primary interaction surface. PHANG systems are designed for this reality, not retrofitted to it.
Trust mechanisms are structural, not cosmetic. Agent verification gates listing creation. Cost disclosure is enforced by the data model, not by policy pages. Complaint routing is wired into the schema. The system cannot be configured to bypass these constraints.