Przemysław Zalewski

For distributors, wholesalers and manufacturers

From the customer’s email to a draft quote, without retyping

Inquiries arrive with PDFs, spreadsheets and scans. Someone splits the items, emails suppliers, waits and retypes prices into a quote. An agent does the reading, the splitting and the chasing, and leaves a draft for your person to check and send.

See price and start date → Book a free 30‑min call

Hourly at $36⁠–⁠54/h excl. VAT. Four to eight weeks, measurement included. Details in the configurator →

Try it first on one inquiry, free: Quote Desk →

Przemysław Zalewski

Sound familiar?

Quote automation problems I fix in 2026

Customers want a price today. Why do we answer in three days?

In Gartner’s March 2026 sales survey, 67% of B2B buyers said they prefer to buy without a sales rep, and 45% had used AI during a recent purchase. An inquiry that waits days for retyped supplier prices loses to the distributor who answers the same afternoon.

I build an agent that reads the inquiry, splits the items and asks suppliers right away, so your salesperson starts from a draft quote instead of a blank sheet. Before it goes live, I measure it on 30 to 50 of your real inquiries. In October 2026 I shipped Quote Desk, which turns a customer’s email, PDF, spreadsheet or handwritten list into lines matched to your price list in about 8 seconds; on 20 synthetic test inquiries 231 of 237 lines were right and it never invented a product code.

Our quoting script sends mail from Microsoft 365 with a password

Microsoft updated its SMTP AUTH timeline: at the end of December 2026 Basic authentication is disabled by default for existing tenants, and in the second half of 2027 Microsoft will announce the final removal date. A script that emails suppliers with a stored password stops sending unless an admin switches the old sign‑in back on, and only until that date.

I connect the agent to your Microsoft 365 mailbox with OAuth through Microsoft Graph, so no password is stored anywhere. I have built IMAP email clients and led Google’s CASA security assessment for OAuth mail access to a pass twice, in January and October 2026.

Can a customer’s email trick our AI into a wrong quote?

On 15 April 2026 researchers showed that one planted line of text hijacks AI agents: a pull request title or an issue comment made Claude Code Security Review, Gemini CLI Action and GitHub Copilot Agent leak their own API keys and tokens back through GitHub. Every inquiry, PDF and scan you receive is written outside your company.

I treat every attachment as untrusted data: the agent extracts items into a fixed format, code checks quantities, prices and product codes, and nothing reaches a customer until a person approves it. In Patriot Mail I designed against prompt injection in email before any model touched a mailbox.

Facing one of these? Book a free 30‑min call →

Scope

What the agent does

Reads the inquiry

Email body and attachments: PDF, Excel, Word or a scan, in Polish, English or German.

Splits the items

By product type or supplier, matched to your own product names and codes.

Asks suppliers

Price and availability requests to the right suppliers, answers collected in one place.

Drafts the quote

A draft in your template or system, waiting for a person to approve it.

Process

How it works

  1. Measure first

    30 to 50 of your real inquiries, and how often items are read correctly, before anything goes live.

  2. Four to eight weeks

    The agent in your mailbox and your system, with suppliers and templates.

  3. People approve

    Nothing goes to a customer until a person approves it.

  4. Then more documents

    Orders, complaints and delivery notes follow the same path.

Where it fits

  • Wholesale and distribution with many items per inquiry.
  • Manufacturing to order, with technical documents.
  • Construction and installation quotes built from drawings and lists.

Track record

Proof

I built a content platform where agents produced thousands of articles and a catalog of 10,000+ programs, ~97% of them found automatically, measured against fixtures. Mailbox access is familiar ground too: Patriot Mail (built at Sanddev), an AI email client, passed CASA Tier 2 twice, in January and October 2026. Count your own case in the free quoting time calculator.

AI content platforms: R&D and an agentic content platform, quality measured not assumed

The problem
Thousands of articles were needed, with AI disclosure built in and without publishing slop.
What I did
Late 2024 I built the agent driver: enforced stop conditions, tool‑call limits, parallelization, delegation chains. Quality came from best‑of‑n, LLM‑as‑judge, scoring against fixtures and plagiarism checks, with many models evaluated across their releases.
Result
Thousands of articles shipped through the platform. Content deployed automatically across 300+ generated sites with git‑backed bulk operations and reputation monitoring. From November 2023 to May 2025 it was a two‑person team. Invented author bylines came off all 300+ sites within a week, and one measured weekend in production saw about 94% of runs succeed at roughly a dollar per article.

thousands of articles 300+ sites LLM‑as‑judge

See also: AI & agents · AI in your company

Affiliate discovery: 300 seeds into a 10,000+ catalog, 97% automated

The problem
A manually curated list of ~300 affiliate programs could not keep up with the market.
What I did
Multi-phase semantic search over pgvector with query expansion and negative queries, embedding-based clustering, tiered caching in Cloudflare R2. Due diligence on 230+ brands against a 50‑criterion scorecard rejected two thirds of them.
Result
A 10,000+ catalog with ~97% of entries found automatically.

10,000+ ~97% automated pgvector + R2

See also: AI & agents · Cloudflare

An AI coding agent as a team member, under enforced rules

The problem
Coding agents write code faster than a team can review it. A team of four wanted the speed without a slop swamp on main.
What I did
I authored the rule set shared by Claude Code and Codex: mechanical checks over written rules, lint and test gates for security, API surface and CSP, agents working on main with hooks, every PR reviewed by a human, AI‑assisted content disclosed as such.
Result
In 2026 about two thirds of merged pull requests across two products were built with coding agents, and nearly all went through my review or merge. I merged 260+ PRs of my own and reviewed 300+ of other engineers’. My commits across those two products went from under 1,000 in 2025 to 7,500+ by October 2026.

~2/3 PRs with agents 300+ reviews 7,500+ commits in 2026

See also: AI & agents · Full‑stack & custom software

Offer: AI engineer for your team (Node.js, LLM agents)

Want the number before the call?

Hourly at $36⁠–⁠54/h excl. VAT. Four to eight weeks, measurement included.

See price and start date → Book a free 30‑min call

Good to know

Reading inquiries with AI: limits worth knowing

Excel and Word need converting first

Claude’s document blocks do not accept .xlsx or .docx; convert them to text, CSV or PDF before sending (PDF support).

32 MB and 600 pages per request

One request carries at most 32 MB and 600 PDF pages, or 100 pages when the context window is under 1M. Password-protected PDFs are not supported (PDF support).

Each PDF page costs 1,500 to 3,000 tokens

Every page is read as text and as an image, typically 1,500 to 3,000 text tokens plus image tokens (PDF support).

The schema does not check quantities

Structured outputs cannot enforce numeric ranges or string lengths, so quantities, prices and codes are validated in code after extraction (structured outputs).

An overnight backlog costs half

The Batch API processes requests at 50 percent off, with most batches done in under an hour (batch processing).

Long prompts on Haiku 5.5 cost more

On Haiku 5.5, a prompt over 100,000 tokens pays higher prices even when part of it is a cache hit (pricing).

FAQ

Questions

Can I try it on my own inquiries first?
Yes, free: Quote Desk reads an inquiry (email, PDF, Excel or a scan) against your price list and shows a draft. The project connects the same thing to your mailbox, ERP and suppliers.
What if the agent reads an item wrong?
The draft goes to a person first, and the share of items read correctly is measured on your documents before launch and tracked after.
Do our documents leave the company?
They are processed by the model provider you choose, on your company’s own account, or by a model on your own hardware if they must stay in house. Before the first real inquiry is processed we sign a data processing agreement (Article 28 GDPR), and the measurement can start on inquiries with customer details removed.
Does it work with our ERP?
Prices, stock and the quote itself can be read from and written to the ERP if it has an API or a database we can use.
How long does it take?
Four to eight weeks including the measurement on your inquiries.
What does it cost, and what does the AI cost to run?
The work is hourly, 36 to 54 USD excl. VAT. Model usage is billed to your company’s own provider account with a monthly spending limit set before launch, and the measurement on 30 to 50 of your inquiries shows the cost per inquiry before you decide on the rest. The configurator shows the range and the start date.
Can AI read the Excel and Word attachments in customer inquiries?
Yes, after conversion: Claude’s document input does not take .xlsx or .docx directly. PDFs work up to 32 MB and 600 pages per request (PDF support).
Can the AI get quantities or prices wrong even with a fixed output format?
The format is guaranteed, but numeric ranges are not (structured outputs). That is why every number is validated in code and a person approves the draft quote.
Customers want a price today. Why do we answer in three days?
In Gartner’s March 2026 sales survey, 67% of B2B buyers said they prefer to buy without a sales rep, and 45% had used AI during a recent purchase. An inquiry that waits days for retyped supplier prices loses to the distributor who answers the same afternoon. I build an agent that reads the inquiry, splits the items and asks suppliers right away, so your salesperson starts from a draft quote instead of a blank sheet. Before it goes live, I measure it on 30 to 50 of your real inquiries. In October 2026 I shipped Quote Desk, which turns a customer’s email, PDF, spreadsheet or handwritten list into lines matched to your price list in about 8 seconds; on 20 synthetic test inquiries 231 of 237 lines were right and it never invented a product code.

In 30 minutes you will know if we are a fit

You bring the problem, I bring the questions. Want a number first? The configurator shows hours, a price range and the start date.

Book a free 30‑min call Email me instead

or write to ciunkos@gmail.com

After you book you get a calendar invite. Nothing else to fill in.

My pledge

  • Reply within 24 hours. I reply within 24 hours on working days, personally, not a bot. Weekend messages: by Monday noon.
  • Late means free. If I am late, the first hour of work is free. You do not have to ask.
  • Scope in writing within 24 hours. A free 30‑minute call with no company deck, then the scope and the price in writing within 24 hours. You decide.
  • The person who does the work. You talk to the person who does the work. No sales layer, no handoff.
  • Follow‑up only with news. I follow up only when I have something new for you. Never a “did you see my offer”.
  • Honest “not a fit”. If I am not the right person, I say so on the call and point you to someone who is.

Calls this week · work can start now · B2B contract · Wrocław, Poland (CET), remote

You own the code, domain and accounts from day one. Nothing is held hostage.