“Phased AI engagements. A one-week paid discovery, then a four-week first phase that ships real value. Written specs for what comes next — instead of twelve-month roadmaps on day one.”
Thirty Birds Labs builds applied AI for organisations that want shipping, not pilots that stall. Every engagement starts with a paid discovery and a tight first phase — deterministic where the rules are clear, model-driven where genuine judgment is needed, chosen on the merits rather than the trend.
Built for shipping, not for the deck — applied product work, phased delivery, and operations automation.
Production AI applications built end to end — from workflow analysis through deployed software with observability, admin controls, and handover documentation. Built to be operated by the client’s team after delivery, not to be a black box.
Engagements run in phases. A paid discovery scopes the problem against real data and real workflow. Phase 1 ships a tight, defensible piece in four weeks. Each phase produces a written specification for the next — so commitments are made on evidence, not slides.
Email triage, document intake, OCR with domain-specific classification, follow-up sequences, mission tracking. Built deterministically where the rules are clear, with models in the loop where genuine judgment is needed. Human-reviewed before anything reaches a client.
Understand the workflow, ship something real in four weeks, scope what comes next from evidence. No twelve-month roadmaps, no shortcuts, no hand-waving — just careful, principled work that compounds.
Every engagement starts with a paid discovery, typically one week. Shadow the team, map the actual workflow, look at real data, and identify the highest-leverage place to start. Sometimes the best answer involves no LLM at all — and naming that early is part of the value.
Map the technical landscape, choose deterministic patterns where they fit and model-driven ones where they don’t. Build for the team that will operate the system after delivery — not just for launch day.
Tight loops against real data and real users. A defensible piece of Phase 1 lands inside the four-week window — validated continuously, not at the end. If a week reveals a problem worth changing course on, that’s the discovery doing its job.
Launch is the start. Delivery includes handover documentation, observability, and a written Phase 2 specification — so the next phase is scoped from evidence, not from hope. Continuing the engagement is a choice the client makes from a stronger position than they started in.
Senior software engineer at Meta, 2018–2024. CTO at Hypeo AI, summer 2025. Five years as tech lead at Dialpad AI (San Francisco) before that, after engineering roles at Dish, Broadcom, and Marvell. UCLA — MS Electrical Engineering, BS Computer Engineering.
Thirty Birds Labs operates as a single-operator consultancy. Engagements are led personally; any subcontracting is disclosed to the client in advance.
LinkedIn →Active engagements as of mid-2026. Confidential details withheld; sector and shape only.
A live WhatsApp agent operating from the CEO desk of an industrial holding in Casablanca. Reference-tracking against a Postgres backend, structured handoffs, and follow-up automation — deterministic flows with model fallback. Live and in daily use.
Internal tooling for a Casablanca chartered-accounting firm: client-document intake and a reusable client-discovery process the firm runs on its own engagements. Scoping the first phase from real workflow.
A four-week Phase 1 scoped against a shared Microsoft 365 mailbox: 5–8 category model, daily digest, human-reviewed auto-drafts. Phase 2 specification delivered as part of Phase 1. No closed-loop sends without human review.
If your team is sitting on a workflow that AI ought to make better, a paid one-week discovery is the lowest-risk way to find out. Then a four-week first phase if both sides see fit. Written specification at every step. No twelve-month commitment.