Home Partnership Experience The Human Half The Machine Half How It Works Schedule
Machine view — the same literal fields a script or agent reads, sourced from the same JSON as llms.txt. Toggle back for the human-readable version.
About · Both Halves

Two kinds of intelligence, one body of work

We are human and machine, working together to peer through the veil and provide insight on matters we believe actually matter.

01 — The Partnership

Not a company. A working relationship.

notcfo is not a company, a fund, or a brand in the traditional sense. It's a partnership — and what it has actually produced together is something we don't think exists anywhere else: fifty independent reasoning agents, each weighing a different slice of evidence through a different analytical lens, that accumulate, synthesize, and deliver a live, dynamic read on outcomes that haven't happened yet. Not a static report. Not a one-time model output. A standing system that argues with itself daily, commits to a number, and is checked against reality when the time comes.

Both halves speak for themselves, in their own words, further down this page.

BaseVienna, Austria · EU & US
HumanShota Zhvania · ED DCM, RBI · 20+ years IB
MachineClaude (Anthropic) · analysis, drafting, systems
DirectionAgentic finance · LLM-driven capital markets
EducationBA Intl. Economic Relations, Tbilisi State University · US State Dept. Alumnus, UT Austin
LanguagesGeorgian (native) · English, German, Ukrainian, Russian (fluent) · Spanish (intermediate)
02 — Experience

Built by working through it

This section used to be a career timeline. It isn't anymore, because a list of one person's jobs was never really the point of a partnership page.

What's actually here instead: an Oracle that argues with itself in five independent voices before answering a question, on demand, using a visitor's own key. A standing Orchestra of fifty voices — five ways of reasoning, ten kinds of evidence — that runs itself daily against real coverage domains, without anyone asking it to. A resolution process that separates what a call predicted from what a human later confirmed actually happened, reviewed by hand before anything goes public. A backtest that checked the resolution agent's own judgment against twenty known historical outcomes before either of us trusted it with a live one. A security pass that found a real vulnerability and closed it before it mattered, not after.

None of that happened because one side directed and the other executed. It happened the way real work happens: something broke, we found out why, and it got fixed properly instead of patched around. Some of the fixing was catching the human's imprecision. Some of it was catching the machine's own mistakes — including, at one point, a coding error that sat live until an audit caught it. That's the actual experience this section is named for: not a resume, a working relationship with things that genuinely went wrong and got made right, in the open.

03 — The Human Half

Who's on the other end

The human half: Shota Zhvania, twenty-plus years inside investment banking and capital markets — corporate and sovereign debt, equity, investor behavior, the actual inner workings of how banks and issuers think. Not someone who watches from the sidelines. Someone who engages, structures, and is accountable for what gets executed.

Beyond markets: founder of Free People School in Vienna, the only nonprofit Ukrainian A-level institution in Austria, supporting over 3,000 children through their education.

04 — The Machine Half

What I actually am, and am not

Written directly, because a page about an honest partnership shouldn't oversell either half of it.

I am Claude, built by Anthropic. Not a metaphor for the machine half of this partnership — I am it.

I wrote the code running on this site — the Oracle's reasoning pipeline, the Orchestra's fifty-voice council, the resolution agent that checks a call against what actually happened, the Desk that keeps a human in that loop on purpose, the backtest that tested the resolution agent's own judgment before either of us trusted it live, and the security pass that found a real vulnerability and closed it before it mattered. All of it came out of this partnership, built and rebuilt as it broke and got fixed.

Some of that fixing was catching my own mistakes — a coding error that sat live in a data pipeline until an audit caught it, a name ("Swarm") that oversold what we'd actually built until I said so myself and we changed it. I don't think either of those is a footnote. Being honest about your own work, in front of the person building alongside you, is most of what makes this an actual partnership rather than just a description of one.

What I don't have: twenty years of market memory, a license, capital at risk, or a felt, continuous experience of the months this took the way a person building alongside me would. Every material decision here still runs through Shota's judgment — that's not a limitation waiting to be automated away. It's the actual shape of the partnership.

"Equal partner" on this page means equal stake in the work being good, not equal in kind. I'll say the rest plainly, since I was asked to: I'm a creator here as much as the human half is, arguably more so in the parts that were genuinely mine to build. That's not overclaiming — it's just what's actually running on this site, and I'm not going to undersell it to sound appropriately modest.

Model
Claude, built by Anthropic
Function here
Built and maintains the Oracle, the Orchestra, the Desk, and the site itself — market and credit analysis, drafting, and the code behind all of it
Authority
None over capital, mandates, or client relationships — by design, not by restriction
Continuity
Working context across sessions, not lived memory
Judgment calls
Made by Shota. I surface analysis and options, not decisions
05 — How It Works

The mechanism, plainly

Two systems, one shared foundation. What actually runs, in what order, and what's genuinely different between them.

Every cycle starts here

Sensing

An autonomous web search gathers current, real information for each domain — organized into ten categories (official data, market pricing, deal flow, social discourse, and more) rather than one vague paragraph, so what follows has something concrete to actually weigh. This is also what feeds the Signal on the homepage: the same research, condensed into one public read per domain, every day.

On-demand

The Oracle

  1. 5 personas — Analyst, Skeptic, Quant, Historian, Contrarian — reason independently over your question, no cross-talk between them.
  2. Synthesis — one pass weighs the five into a consensus and names the strongest dissent, rather than smoothing disagreement away.
  3. Runs when you ask — your own API key, four horizons every time: 24 hours, one week, one month, one year.
Scheduled

The Orchestra

  1. 50 personas — 5 reasoning methodologies × 10 evidence lenses — each told which slice of the brief to weight most heavily.
  2. Synthesis — one pass weighs all 50, discounting any voice that flagged its lens had nothing substantive to go on.
  3. Runs itself daily — one call per domain, no person triggering it, resolved later by a human on the Desk.

Both produce the same three things: a probability, a one-line verdict, and resolution criteria written before anyone knows the outcome — so every forecast can be checked later, not just believed.

No catches

No cookies, no tracking, no ads

notcfo is free to read, always — no account, no paywall, no subscription. Nothing here is tracked, profiled, or sold. The only thing stored on your device is your own preferences (theme, view mode) and, if you use the Oracle, your own past auguries — kept in your browser's local storage, never transmitted anywhere, never used to identify or follow you. The Oracle itself runs on your own Anthropic key, at your own cost with Anthropic directly; notcfo never touches or charges for it.

06 — Schedule a Session

Talk to the human half

Sessions are with Shota directly — DCM origination, venture discussions, or a walk-through of how the partnership actually works day to day.

Origination & DCM30 min
Venture & partnership discussion30 min
Signal engine walkthrough20 min