Victor Algaze — AI Engineer — Hermosa Beach, CA

Your founding
AI engineer.

One engineer, 275+ portfolio companies of leverage. That’s the trade. I build production AI systems that do real jobs — and I grade them the way you underwrite deals: against reality, not demos. I don’t do slideware. I deploy.

Victor’s Agent Verified facts only

Ask me about his background, his shipped work, the K1 plan, or his stack. It will not invent facts. If it doesn’t know, it says so.

Victor Algaze, smiling in front of a whiteboard
Victor AlgazeHermosa Beach, CA

That’s me, mid-build. That whiteboard has watched me talk myself into bigger projects than this one.

For the founding partner

This page is the demo.

You said sitting through the product demo has never been more important. So here it is: production systems, real users, measured outcomes. No slideware.

Your bar: green/yellow/red on depth of workflow, system-of-record status, and uniqueness of data. I answer it in green/yellow/red below. And I grade my own work the same way — a 14-code weighted rubric, expert review.

Claude is my daily tool: multi-step agent workflows, model evals, written-up recommendations. Smarsh — a K1 company — ships a Claude Compliance API integration you reposted. MariaDB is shipping vector search for AI workloads. Your portfolio is already on my stack.

Two tours at Cisco. CTAO 2016–2019: IoT device management, the $ugar internal library, collaborations with Disney and the City of Paris. CX Engineering Incubation 2020–2024: a conversation assistant from first build to live production on CX Cloud. Back since March 2026 via PortalCo on the agentic-AI program. UC San Diego 2005–2009, B.S. Management Science. 2019 Cisco Young Pioneer Award. The unglamorous part is shipping.

Your words: two foundational LLMs today — “within a year, I think it’ll be one if not sooner.” That call is coming. I’d help make it.

And the thesis: “single team, single office, single strategy.” One engineer on this seat is leverage across every company you touch.

Victor Algaze · Hermosa Beach · ~15 min from Manhattan Beach

The screening question, answered

Diligence intake, automated.

Your application asks: “Please share one process you’d automate with AI and how you’d build it.” Here’s my answer:

It sits where the money leaks: diligence quality and deal-team time.

1.

Ingest

An agent reads the whole data room. CIMs, financials, customer contracts. Every document, every spreadsheet.

2.

Extract

KPIs land in K1’s investment-memo scaffold — every value cited to the source document and page.

3.

Reconcile

Revenue in the CIM vs. the audited financials vs. the board deck. Every figure that doesn’t tie across documents gets flagged.

Step 4: the one that matters

A human approves.

Nothing reaches a partner without a person signing off. The output is a pre-drafted memo scaffold: the KPIs, the discrepancies, the source trail. The deal team starts from a draft, not a blank page.

Measured like an underwriter, not a demo: extraction precision/recall, discrepancy catch rate, hours saved per deal. Green/yellow/red on the workflow — never vibes.

Then: board-report generation, competitive-intel scrapers, measured model evals — the whole job description, iterated against real user feedback. Rapid tools and agents, dashboards, data pipelines. Deployed, not demoed. One version of the truth — that is a moat.

The operation

AI tools, in production.

29 Court is an AI operation: agents doing real jobs, with a daily briefing on top. Here’s what’s live.

Live

AI for an expert-witness psychologist

A HIPAA-compliant LLM system for Healthcare AI Solutions. Real deployment, real compliance constraints.

Jul 2025 — present

80%
Report time cut
Live

Cisco’s agentic-AI program

Multi-tool agent workflows, orchestration, and eval harnesses. The day job, run through PortalCo while I run 29 Court.

Mar 2026 — present
Live

AutoCAD plugin + cloud infra

BOMA floor plans for Realm8, plus the cloud behind it. Boring reliability.

Feb 2025 — present
Shipped

Email grader, built from scratch

14 weighted codes, a 0–100 composite score, and a review loop for domain experts. It grades. It doesn’t flatter.

0–100 score · expert review
Shipped

Conversation assistant, CX Cloud

From first build to live production. The exact 0→1 muscle a founding team needs.

2020 — 2024
Private beta

29 Court Platform

The operation’s home: platform.29court.com — “Your agents, run like infrastructure.”

platform.29court.com
Private beta

Armadillo

A private-beta backend toolkit on Cloudflare. The boring plumbing (users, auth, data, jobs) my agents run on.

Cloudflare-native
The cadence

Every job ends in a briefing. What the numbers say, what to watch, the recommended call. That’s what I’d put on a founding partner’s desk every morning.

Against the bar

Green, yellow, red.

Neil Malik triages software on three things: depth of workflow, system-of-record status, uniqueness of data. My read on myself, against his bar:

Depth of workflow

My agents live inside the workflow — intake, extraction, reporting — not alongside it. And I’m in the room where the work happens, five days a week.

Green

System-of-record status

Outputs land in your formats: the investment-memo scaffold, the board-report template. Human-approved before anything reaches a partner. The system of record stays sacred.

Green

Uniqueness of data

Every deployment compounds into measured data: extraction accuracy, discrepancy catch rates, hours saved per deal. Each one makes the next one sharper. The moat compounds from deployment one — building it is the job I’m applying for.

Yellow

Two green, one honest yellow

The stack

The shipped stack.

Runs on
Python · TypeScript · Cloudflare Workers
Daily driver
Claude: multi-step agent workflows, custom eval harnesses
Ships
Agentic workflows · data pipelines · dashboards
Measures
Model evals, written up as recommendations

Contact

Talk to me.
Or talk to the agent.

Five days in-office, fifteen minutes away. The fastest way to evaluate me is the chat at the top of this page. The second fastest is one of these:

valgaze@gmail.com↗ valgaze.com↗ 29court.com↗