Website · NOVA Labs · LinkedIn · Email
Amir Hossein Kazemkhani — AI systems engineer and founder of NOVA Labs.
I build voice agents that operate in real time, evaluation systems that produce evidence,
and developer tools that make the hard parts reusable.
|
5 use-case suites · 36 controls · Arabic / English An adaptive assurance engine that produces append-only SHA-256 evidence artifacts and inspectable Arabic / English assurance records for public-service AI. |
12 MCP tools · 47 tests · v1.0.1 A typed server that puts voice-agent calls, live status, quality, usage, and webhooks directly inside MCP-compatible AI clients. |
|
409 tests · gated sequencing · 0 release-audit findings A TypeScript control plane for explainable ICP scoring and multi-channel sequencing, with every real action behind an explicit human approval gate. |
Live deployment · 4 showcased systems · technical writing The deployed public home for my shipped systems, engineering record, field notes, and work at NOVA Labs. |
Note
Proof over theatre. phantom-transition isolates a subtle failure in interruptible voice agents: a reply can be cancelled while its tool call survives. The repository reduces it to a zero-runtime-dependency model and a 24-test guard suite.
OBSERVE the real failure
└─▶ CONSTRAIN the system boundary
└─▶ PROVE the behavior
└─▶ PUBLISH the reusable part
- Voice agents — low-latency conversation systems, tool use, interruption safety, and the operational layer around them.
- Evaluation infrastructure — repeatable probes, explicit decision rules, evidence bundles, and deployment gates.
- Open developer tools — focused packages and reproductions that turn production lessons into inspectable code.
Building evaluation infrastructure for evidence-backed AI decisions, strengthening NOVA's real-time voice stack, and publishing the small tools that make agent systems safer to operate.
Dubai, UAE · Build the system. Measure the behavior. Show the evidence.
Explore the work →