Consulting

I help senior engineering teams ship reliable multi-agent systems in Python.

No fake-green test theatre, no agent demos that fall over on second use. Forward-deployed engineering on your repo, on your stack, with a deterministic orchestrator (Bernstein) underneath so the work is reproducible after I leave.

Things shipped

Bernstein — open-source governance layer for AI agents. Deterministic scheduling, HMAC-chained audit trail, per-agent credential scoping. Apache-2.0, 40+ CLI adapters. Used to orchestrate fleets of 10–2000 agents across coding, research, and operational tasks.
Model routing and provider resilience — a routing layer across 100+ models from multiple providers: failover, per-tenant rate limits, cost accounting. The layer Bernstein crews run on.
GlyphShift — macOS utility that detects when you type in the wrong keyboard layout (RU/EN/HE) and retypes the word in the right one. On-device CoreML, no cloud.

Shipping these systems today, not presenting decks about them.

Engagement

If your team runs multi-agent systems in production and needs them governed, audited, or scaled — one call, then a scoped proposal.

Two lines before we talk, so I can prepare: the problem in one line, plus team size and domain.

Book a call or email directly

What working with me looks like

Architecture review — one week. For teams whose AI system is nearly live and needs clearer seams and less accidental chaos.
Bernstein deployment — two to four weeks. Stand up a multi-agent crew on your repo with file-based state, per-agent credential scoping, signed audit trail.
Embedded build shaping — two to three months. I join your team as a senior engineer and shape the workflow, the system boundaries, and the path to production.

Case studies

Redacted write-ups from prior engagements are shared on request after a first call, under NDA.

Reading before the call

Things shipped

Bernstein — open-source governance layer for AI agents. Deterministic scheduling, HMAC-chained audit trail, per-agent credential scoping. Apache-2.0, 40+ CLI adapters. Used to orchestrate fleets of 10–2000 agents across coding, research, and operational tasks.

Model routing and provider resilience — a routing layer across 100+ models from multiple providers: failover, per-tenant rate limits, cost accounting. The layer Bernstein crews run on.

GlyphShift — macOS utility that detects when you type in the wrong keyboard layout (RU/EN/HE) and retypes the word in the right one. On-device CoreML, no cloud.

Shipping these systems today, not presenting decks about them.

What working with me looks like

Architecture review — one week. For teams whose AI system is nearly live and needs clearer seams and less accidental chaos.

Bernstein deployment — two to four weeks. Stand up a multi-agent crew on your repo with file-based state, per-agent credential scoping, signed audit trail.

Embedded build shaping — two to three months. I join your team as a senior engineer and shape the workflow, the system boundaries, and the path to production.