Taking new engagements · written quotes within 3 business days
AI integration for finance

AI should never touch money unchecked.

I build the layer between AI models and your brokers, portfolios and internal data. The model reads, reasons and drafts. Code checks every number, a person approves every action, and every decision is signed and logged.

Written scope and fixed price within 3 business days. No sales calls.
100% Job Success across 101 client projectsverified on Upwork ↗
Any model ClaudeGPTGemini Llama · Qwen · Mistral, self-hosted
rebalance-review · mock dataidle
A live illustration on mock data. Try approving or rejecting the proposal.
Approach

Where AI in finance goes wrong, and what I build instead.

Getting a model to answer questions is easy. The work that matters is everything around it: who does the maths, who can act, what gets recorded, and where your data goes.

The usual way
How I build it
01

The model adds up the portfolio.

Language models guess at arithmetic. A transposed digit in a rebalance is a real trade.

The model never does the arithmetic.

Weights, drift and exposure come from deterministic code. The model explains the numbers; it doesn't produce them.

02

The agent places the order.

One confident mistake and it has already reached the broker.

Nothing moves without a human.

Orders, alerts and client messages stop at an approval step the model cannot skip.

03

Nobody can say why it did that.

When a regulator or a client asks, there's a chat transcript at best.

Every decision leaves a signed record.

What the model saw, what it proposed and who approved it, in a tamper-evident log you can replay a year later.

04

Client data goes to someone else's API.

And the vendor choice gets made before anyone has measured anything.

Your data stays where you decide.

A cloud model under your own account, or an open-source model hosted inside your network. Chosen on your own evaluations.

Work with me

Built for three kinds of finance teams.

Fixed scope, fixed price, delivered into your own infrastructure with tests and documentation. You talk to the engineer who builds it, from the first message to handover.

Advisors & RIAs

Private AI for your advisory practice

Turn custodian exports, spreadsheets and research into workflows your advisors trust, without client data leaving your control.

  • Meeting briefs and quarterly review drafts
  • IPS drift checks computed in code
  • Watchlist and filing digests
  • Built on sandbox data; you connect live credentials
Includes
Up to 3 workflows · 2 data sources · 1 environment
Add more
+$2,500 per workflow · +$3,000 per data source
Warranty
30 days of free fixes
From $10,000
3–4 weeks
Funds & family offices

Private research and portfolio intelligence

Connect AI to your databases, feeds, filings and multi-account positions, with your signals never leaving your walls.

  • MCP servers over your internal systems
  • Cloud or self-hosted model, chosen on your evals
  • Sizing and risk checks in deterministic code
  • Approval gates and signed decision logs
Includes
2 internal systems · 3 workflows · 2 models compared
Add more
+$4,000 per system · +$2,500 per workflow
Warranty
30 days of free fixes
From $15,000
4–6 weeks
Fintech & trading platforms

Your official MCP server

Let your users work with your platform from Claude, ChatGPT, Cursor and other AI clients, safely.

  • OAuth 2.1 and per-user permission scoping
  • Clean, typed tool contracts
  • Approval-gated actions for anything that trades
  • Automated evals in CI, so tools don't regress
Includes
1 API · up to 12 tools · OAuth · CI evals · docs
Add more
+$3,000 per 5 tools
Warranty
30 days of free fixes
From $15,000
3–6 weeks
Not sure yet?

Start with a pilot.

One workflow over one data source, built end to end on sandbox data in 1–2 weeks. $3,000 if it drafts and reports; $5,000 if it also takes an approval-gated action. You see exactly how I work on your problem before committing to the full build.

$3,000–$5,000credited toward the full buildAsk for a pilot quote

After delivery, an optional monthly retainer keeps it working as broker APIs, MCP and models change.

How it works

From your first message to a fixed price in three days.

01

Send your requirements

What the AI should do, the systems involved, your constraints and timeline.

About 5 minutes
02

Get a written quote

Scope, architecture outline, fixed price and timeline. Or an honest note that it isn't a fit.

Within 3 business days
03

Fixed-scope build

Weekly working demos on sandbox data. You see it running long before it's finished.

3–6 weeks
04

Handover

Deployed in your environment, with tests, documentation and a walkthrough for your team.

Then optional support
Fixed price, clearly bounded

What the price covers.

A fixed price only works if both sides know exactly what's in it. So it's written down before any work starts.

  1. A written scopeThe exact workflows or tools, the systems they connect to, and the tests that decide when each one is done.
  2. Changes are quoted firstAnything new gets a written price before I start it. You approve it, or move it to a later phase. No surprise invoices.
  3. Paid in two parts50% to start, 50% when the acceptance tests pass. Larger builds split into milestones.
  4. 30-day fix warrantyIf something I built doesn't do what the scope says, I fix it at no cost for 30 days after handover.
Fixed price per scope. No hourly meter.
Built on sandbox data. I never handle your client PII.
You own the code built for you.
Engineering only. No investment advice, no promised returns.
Proof

Don't take my word for it. Check.

Open code you can read, client reviews you can verify, and a decade of engineering where mistakes were not an option.

Client work stays confidential. I'm glad to walk through architecture and trade-offs in a paid session.
Writing

Notes on building AI that handles money.

Short, practical teardowns of the decisions that make or break AI systems in finance.

Fit

I'm selective, so you know what you're getting.

A good fit

  • RIAs, family offices and funds putting AI on research or portfolio work
  • Fintech and trading platforms shipping an MCP server to their customers
  • Teams that want it built properly once, not prototyped three times

Not a fit

  • Trading bots sold on promised returns
  • Investment advice, signals or strategies
  • Hourly staff augmentation
Questions

Before you ask.

Which AI models do you work with?+

Cloud models such as Claude, GPT and Gemini under your own enterprise account, or open-source models hosted inside your network when data can't leave. I recommend one based on evaluations against your own tasks, not loyalty to a vendor, and the tools I build work across models.

Will you see our client data?+

No. I build and test on sandbox or anonymized data, then deploy into your environment. Your team connects the live credentials.

Do you give investment advice or trading signals?+

No. I build the systems your team uses. Decisions, strategies and results stay yours, and I never promise returns.

How is pricing set?+

A fixed price per scope, quoted in writing before any work starts. Builds start at $10,000. If you'd rather test the fit first, a pilot of one workflow costs $3,000–$5,000. If the scope is too unclear to price, start with the paid architecture session. Both are credited toward the full build.

What if we need more, or less, than the quote covers?+

Each offer lists what's included and what extra workflows, systems or tools cost, so you can size it yourself. If you need more mid-build, I quote the change in writing before starting it. If you need less than a full build, start with a pilot. And if a request is too big for one quote, I split it into phases, each with its own fixed price.

What do you need from us?+

Sandbox or test access to the systems involved, one person who can make decisions, and feedback within two business days. If access or feedback is delayed, the timeline moves; the price doesn't.

Who owns the code?+

You own everything built for you. The contract names any general tooling I bring from before, so there's no lock-in.

Can our compliance team review it?+

Yes. You get a written description of what the system can and can't do, where approvals sit and what's logged, alongside the code and tests.

Where are you based?+

Cairo, working with clients in other regions. I work your hours, and engagements are contracted through my company, AK Trading.

Request a quote

Tell me what the AI should do.

You'll get a written scope and fixed price within 3 business days, or an honest note that it isn't a fit.

  • I read every request myself.
  • If something is unclear, I'll ask by email, not book a call.
  • The quote covers scope, architecture outline, price and timeline.
Prefer to talk it through?Architecture session · 60 min · $300, credited toward your build
Book a session
Who will use it?
Full builds start at $10,000. A pilot ($3,000–$5,000) or the paid architecture session may be the right first step, and both are credited toward the build.
Sent to me directly. Never shared or added to a mailing list.