Trust-critical AI, built by Logan Chandrasegaran.
Three specialist agents over shared memory, in an autonomous build → verify loop. The model drafts — it never decides.
By Logan Chandrasegaran, Founder & AI EngineerWhat Blimp builds↓Ask me about Blimp.
frank, the operator · grounded · says when he doesn't know
The system
Not a monolithic prompt — a real topology. Each agent owns a role and its own runtime; they coordinate over shared persistent memory.
Hermes leads architecture, Duke leads the build, Frank runs operations — each its own runtime, not a monolithic prompt.
Agents coordinate over shared persistent memory. Context compounds across sessions.
Built, then verified before it ships — under a codified constitution.
What Blimp builds
AI co-pilot for employment-law cases
Plain-English guidance for Malaysian HR teams on disciplinary and dismissal cases, grounded in real Industrial Court awards.
The hard problem solved
A deterministic citation-integrity gate verifies every citation programmatically — the model is never trusted to cite.
Customs-duty HS-code classifier & landed-cost engine
The LLM classifies; pure functions compute landed cost over versioned tariff tables. Numbers are never left to the model.
Low-latency conversational AI
mic → WebRTC → Deepgram → Claude → Cartesia, wired into persistent memory. End-to-end verified.
How Blimp builds
Creative work to the model; consequential work to code.
Numbers, citations, and rules computed by pure, testable functions — never the LLM.
Every output scored for faithfulness and provenance before it ships.
Guardrails and codified governance from day one, not bolted on.
Capabilities
Track record
Contact
Open to senior AI-engineering roles. Kuala Lumpur, Malaysia (GMT+8).