Full-Stack & Geospatial Engineer · Agentic Systems Builder
Founder @ Khamseen Technologies

I build production systems at the intersection of software engineering, geospatial platforms and AI orchestration.

• Agent infrastructure — control planes, capability delegation, Skills, runtime contracts, policy enforcement and audit.
• Geospatial platforms — ArcGIS Experience Builder, ArcGIS Maps SDK for JavaScript, spatial workflows and API-driven GIS applications.
• Full-stack systems — TypeScript/React, Python/Django, PostgreSQL, Redis and containerized infrastructure.
• Reliability — deterministic checks, independent QA, idempotency, execution leases, recovery and observable costs.

An agent-native enterprise control plane: one human authority coordinating specialized agents, deterministic engines and replaceable execution providers.

Khamseen owns intent, authority, Task/Run/Unit state, policy, approvals and audit. Execution infrastructure plugs into those contracts. The direction is a modular Core supporting Commerce, Career, Finance, GIS, SaaS and R&D through scoped business modules.

Graphs coordinate. Loops converge. Dependencies reflect actual data and state consumption; independent Units can run in parallel. Each Unit uses a bounded produce → check → correct loop. Atomic checkout prevents conflicting claims, while execution leases and heartbeats support explicit orphan recovery.

Capability ≠ Skill ≠ Tool. Authorization, procedural knowledge and external actions stay separate. The Skills direction uses versioned SKILL.md contracts and progressive loading: metadata first, procedures when needed.

The plumbing stays replaceable. RuntimeAssignment and HarnessProvider separate agent identity from execution. Hermes, OpenClaw, Cursor, Codex and Claude Code belong behind this boundary; OpenClaw remains an evaluation track. SecureRuntimeProvider / SandboxProvider define isolation, while durable execution handles checkpoints and resume without owning domain state. OpenShell, Daytona, Inngest, Temporal and agentgateway are candidates whose adoption must justify a distinct role.

Context ≠ Memory ≠ Knowledge ≠ Policy. Scoped context serves the current Unit. MemoryProvider handles experiences and observations; KnowledgeProvider handles documents and cited retrieval. Hindsight, WeKnora and PageIndex are evaluated behind those boundaries. Context Mode is an optional context-optimization candidate. Learning becomes policy only through aggregated evidence, validation and an explicit gate.

Review produces evidence. Aegis evaluates it. Deterministic checks precede model judgment. ReviewProvider can supply specialized findings, with OpenCodeReview under evaluation; independent review remains a separate responsibility. Implementation, correction and review produce distinct execution evidence.

I study Paperclip for control-plane patterns, ECC for selected Skills and security evidence, Archify for architecture artifacts, Impeccable for UI quality and Univer for Office outputs. These are evaluation or reference tracks, not a list of shipped integrations.

The current priority is V0 closure → Skills foundation → Graph / Unit Runtime, with secure execution, economics, memory/knowledge, learning and harness portability developed around their actual dependencies.

Human involvement moves toward goals, exceptions and consequential approvals. Authority and evidence remain explicit throughout.

_{Autonomous execution. Explicit authority. Verifiable outcomes.}