SUNEB turns frontier AI into systems teams can actually operate.
We package models, workflow automation, agents, and vertical products into measurable business outcomes: hours saved, deals closed, students retained, and operations controlled.
AI operations for Nepal-first organizations with global ambition.
- Primary base
- Kathmandu, Nepal
- Operating mode
- Audit, build, tune, hand over
- Core protocol
- UNCERTAIN-stop
- Commercial path
- Starter pack to managed system
The UNCERTAIN-stop protocol.
SUNEB does not let agents invent Nepal law, fake government APIs, or pretend confidence around Devanagari text. Unverified work stops, records the reason, and waits for human review.
// SUNEB agent output contract
Operating philosophy
SUNEB is built around useful control: know the process, name the risk, ship the smallest system that can prove value.
Nepal-first context
Local business patterns, education workflows, language realities, and compliance questions are treated as product requirements.
World-capable build quality
Systems are scoped, tested, documented, and handed over like production software, not one-off demos.
No hallucinated certainty
Agents stop when a claim, integration, or legal reference cannot be verified.
Model-agnostic stack
SUNEB chooses models and tools by risk, cost, latency, privacy, and client reality.
Founded by Sujan Pokhrel.
Based in Kathmandu, SUNEB is focused on the gap between global AI capability and Nepal's practical operating needs. The first-year focus is education AI, SME automation, and agentic workflows that can become long-term products.
Stack with operating judgment.
The stack changes when the client reality changes. Privacy, cost, latency, accuracy, and maintainability decide the tool.
| Tool | Role | Why it exists |
|---|---|---|
| n8n CE | Orchestration | Workflow automation and system routing |
| Gemma 4 E4B | Local AI | Private inference path for sensitive work |
| DeepSeek V3.2 | Reasoning | Cost-aware planning and analysis |
| Gemini Pro | Production | Enterprise-grade inference when scale matters |
| OpenClaw | Agentic framework | Multi-agent orchestration and delivery control |
Names are not decoration. They define responsibility.
Each agent has a division, a role, and a boundary. That makes the operating system easier to explain, inspect, and improve.
Assessment, roadmap drafts, and opportunity scoring
Campaign planning, content operations, and GEO workflows
Sales pipeline intelligence and outbound operations
Brand, video, and content production systems
School workflows, AI literacy, and education automation
Start with the smallest useful next step.
The first conversation should clarify whether you need an audit, a starter pack, a custom build, or a no-build recommendation.