Ibrahim Naeem
Flagship system

Auto Pilot Events OS

An AI operating system that runs the deal lifecycle of an experiential events company: pricing, proposals, sourcing, staffing, finance. It reads the company's own documents, learns what things actually cost, and does the work that used to sit with a senior estimator's memory — in production since 12 July 2026, built and owned solo.

28 days
Concept → live
95
Registered AI tools
7+1
Agents + director
753
Automated tests, green
$0/mo
Running-cost floor

That floor runs on free local and free-tier cloud models — genuinely usable, not a demo mode. For consistently accurate pricing and drafting at real volume, a realistic running cost with paid models is roughly $15–40 a month for a lean setup, scaling to roughly $100–300 at full deployment. Any tier can call in a top-tier flagship model — Claude, GPT, Gemini — for the specific job that needs it: creative generation, complex reasoning, day-to-day operations. The model is a setting, not a rebuild.

Why it exists

Every events company's real intelligence lives in a few heads and old spreadsheets.

An RFP lands. Someone digs through last year's quotations to remember what a stand, an LED wall or a custom counter actually costs. Pricing drifts between estimators. When a senior person leaves, years of pricing instinct leave with them, and nothing the company learned on one deal makes the next deal any smarter.

Auto Pilot Events OS is the fix, built after twelve years of watching that exact problem from the floor: a system that turns every document the company already produces into institutional memory it can act on — so pricing, proposals and planning stop depending on who happens to be in the room.

What it does

Eleven things it does without anyone touching a spreadsheet.

Ranked by how much each one actually changes a working day, most impactful first.

DOCUMENTS IN RFPs · supplier quotes · 3D design files · pricing sheets · chat or voice AUTO PILOT EVENTS OS Reads every document · learns every price · runs the deal Pricing confidence-tiered Proposals seconds, not hours Staffing conflict-checked Finance cash flow, P&L RESULT A decision a person used to make from memory, made the same way every time
  1. Reads any document, learns every price
    Drop in an RFP, a supplier quote, a 3D design file or a pricing sheet — it extracts every priced line item and learns from it automatically. No manual data entry, ever. Every other feature below runs on top of this one.
  2. Prices new work from the company's own history, not a guess
    A machine-learning pricing model trained on 1,136 real historical price points, cutting the typical pricing miss by 66% versus the raw baseline. It only ships a number when the evidence backs it — otherwise it flags the item for a human instead of guessing.
  3. Turns a brief into a client proposal in seconds, not hours
    Internal cost sheets, client-facing PDFs with margins hidden, RFP compliance checks and full proposal packages, generated in the company's own house style. Work that used to be manual formatting is now a request.
  4. Runs the company's finances in real time
    Cash flow projected 13 weeks out, receivables and payables tracked, pipeline revenue forecast by deal stage, a full profit and loss — computed straight from real event data, with no spreadsheet upkeep.
  5. Talk to it — type or speak, it does the work
    A natural-language assistant plans and runs the right chain of tools itself and narrates each step as it goes, with voice input and output built in. No training required to operate a 95-tool system.
  6. Seven specialist agents, each boxed into its own job
    An intake clerk, a pricing analyst, a sourcing scout, a commercial drafter, a render reader, a QA verifier and a data scientist, coordinated by a director. The drafting agent physically cannot send a client file — only queue it. The risk every CEO worries about with AI, an agent acting outside its lane, is designed out at the architecture level.
  7. Builds staffing plans and books resources itself
    Generates crew plans straight from a design's item list, books people and equipment with conflict detection, and turns logged hours into real labour cost — the double-booking that turns into a live-event crisis gets caught before it happens.
  8. Flags the failures that actually sink events, before they happen
    Every deal within 7–21 days of its event date is scanned for missing suppliers, no crew assigned, unsent quote requests or unpaid deposits, and the whole business rolls up into a single 0–100 health score with a transparent reason behind it.
  9. Builds the supplier and client database on its own
    Every quote and RFP ever dropped in becomes a scored supplier and client record automatically. No one sits down to maintain a CRM — it maintains itself.
  10. Runs from free to full-scale on the same system
    One setting moves the whole thing between four cost modes, from a free laptop setup to a scaled cloud deployment at roughly $100–300 a month — same features, same database, nothing rebuilt to grow.
  11. Keeps a full audit trail of who did what — including the AI
    Every action, human or agent, is written to an append-only log: not a bolt-on feature, the architecture itself. Daily backups and tested restores sit on top of it, along with margin floors that block a wrong AI-suggested price before it ships.
The math

Measured on real data, not projected.

The one number I'll put a stake in the ground on, because it came out of a validation run against the company's own historical pricing corpus, not a pitch deck.

Pricing error, median absolute miss
Baseline66% lowerlive-verified
A two-thirds reduction in pricing error, achieved by teaching the system to tell real line items from lump-sum totals and allowance buckets — 57% of the raw training data turned out to be noise, filtered out by deterministic rules, no guesswork involved.

What I won't claim: no client-facing hours-saved or revenue-ROI figures yet. The system has been in production since 12 July 2026; those numbers get published once real usage produces them, measured the same way as the number above.

Twelve years of knowing what breaks is the actual moat. The code is the easy half.
Under the hood

For anyone who wants to kick the tires.

FRONT DOORS document drop zone · RFPs · SketchUp / 3D files · renders · pricing sheets · chat THE BRAIN Typed API · auth & role-based access · 95-tool registry THE SPINE Append-only event log: audit trail · agent memory · training data Agent runtime 7 specialists + a director, 24/7 Learning loop every correction → training signal Control room 23-page web app + chat assistant Model gateway routes every AI call: free local models ↔ free cloud ↔ paid frontier. Swappable, never load-bearing
Event-sourced spine

Postgres with pgvector. Every state change is an event, so the system can be replayed and audited, not just queried.

A real application

A 23-page Next.js front end audited down to 375-pixel phone screens — not an API and a promise.

Built to be trusted

Hard-blocked actions (sending money, sending external messages, deleting data) stay off until a human explicitly lifts them.

Test-gated, always

753 automated tests grown batch by batch, never bolted on after. Nothing ships without the suite green.

The full engineering account — the five design principles, the governance ramp, the four cost modes, and the section on what I deliberately won't claim yet — is in the build story.