Brief
AI that asks first, then checks its own work. Turns a vague request into a precise, gap-checked spec so anyone gets expert-level output on the first try. Sold per seat to teams.
beforenine.ai
Rizqi Ventures builds and ships AI products, and advises the teams deploying them. Fractional Chief AI Officer work, P&L-aligned, with production software when off-the-shelf doesn't fit.
Two-time venture-backed founder. Y Combinator W17. ~15 years building production systems at MuleSoft, Box, and Blameless. Now: building AI products specifically for real estate teams — and embedding part-time as a Fractional Chief AI Officer for teams that want a working AI stack faster than their competition.
Rizqi Ventures LLC builds AI products and advises the teams deploying them. The products below are live and in market.
AI that asks first, then checks its own work. Turns a vague request into a precise, gap-checked spec so anyone gets expert-level output on the first try. Sold per seat to teams.
beforenine.ai
Four areas of work, one role. I show up as your Fractional Chief AI Officer — part-time but accountable. Strategy and governance set the bar; P&L alignment keeps us honest; and when off-the-shelf software doesn't fit, I build it.
Most teams over-invest in tool selection and under-invest in adoption. The playbook below is built around the opposite bias.
Where AI creates margin in your business — not the surface-level feature your board is asking for. Which bets to make, which to defer, what the next 12 months look like.
Every workflow, vendor decision, and build is tied to a specific line on your P&L — revenue, cost-per-side, ops overhead, recruiting cost. Re-reviewed quarterly so the work always points at the numbers that matter.
Brand voice, compliance, data handling, AI literacy training for the leadership team. The unsexy work that determines whether you sleep at night and whether your team actually uses what we build.
When the right answer is custom software — an internal tool, a workflow no vendor sells, an integration between systems that don't talk — I write the code. Production-grade, hosted, monitored. Owned by you on delivery.
Founder twice. Acquired once. IPO twice as part of the engineering and product leadership at MuleSoft and Box. Building Bounti.ai full-time today.
AI agent platform for real estate. Two products in market — Bounti virtual staging (photorealistic staging, declutter, redesign) and B.Claw (always-on AI agent across Gmail, MLS, CRM, calendar, DocuSign, 250+ tools). Vertical: real estate, TAM 2.3M agents.
GV-backed
Category-defining SRE platform. Zero to acquisition. Customers included Procore, Home Depot, Mercari.
$50M+ raised · Acq. Freshworks
Senior engineering and product leadership through MuleSoft's IPO and the $6.5B Salesforce acquisition — one of the largest software acquisitions in history at the time.
IPO · $6.5B exit
Engineering and product leadership from earlier growth stage through Box's IPO on the New York Stock Exchange.
IPO
Everything is month-to-month. No fixed-term lock-in. Travel passed through at cost. Specifics — pricing models, payment paths, build scopes — covered in a short proposal once we’ve had a first call.
P&L review with you + diagnostic across leadership, ops, sales, marketing. Output: written opportunity map tied to P&L lines + 12-month AI roadmap, presented in person.
Capacity · 2 / quarter
Embedded ownership of your AI strategy, execution, and vendor decisions. Monthly on-site, weekly remote, ships workflows live. Custom builds absorbed in retainer hours or scoped as fixed-bid projects on top.
Capacity · 1–2 active
Production software builds for teams that need something no vendor sells. Hosted, monitored, source handed over. Same person doing the strategy writes the code — no vendor handoff.
Capacity · Selective
Working notes more than polished essays. Each is a thread I’ve been pulling at while building production AI products. Full posts go up here as they’re ready.
Most teams ship AI features faster than they ship the evaluations that keep them honest. The bill comes due in production silently — regressions show up as customer emails instead of failing builds.
You don't get good agent behavior by scripting every output. You get it the way you get good behavior from a child — by setting values and boundaries, then letting feedback do its work.
Helpful is the easy one. Harmless is well-studied. Honest is where trust compounds — and where most consumer AI products quietly cheat to keep engagement up.
Agentic loops where flexibility matters. Typed pipelines where reliability matters. The hardest design problem in building real AI products is knowing which boundary you're on.
A prompt tuned on one frontier model quietly falls apart on another. Schema-validated output and a model-agnostic prompt layer are not optional once you route across providers.
The fastest path is email. I read everything, but I reply to short, specific notes first — what you’re working on, what’s in the way, and what you’d want from a first conversation.