>_DietrichDEV
CS

Pavel Dietrich/Forward Deployed Engineer & AI Product Developer

AI that makes it
past the demo.

I embed with your team, take apart the process that slows you down, and build AI software on top of it that runs in production. Wired into your data, your systems, and the people who use it every day.

Discuss a projectSee the work
7+ years
as CTO in fintech
Fundlift Award
Vistag, UP21 accelerator
.NET · TS · LLM
from database to model

01

Services

I take responsibility for outcomes, not billed hours.

  1. 01

    Forward deployed engineering

    I don't work through tickets from across the country. I sit with your team, next to the people who run the process every day, and we decide together what's worth automating and what isn't.

    • on-site discovery
    • prototype on real data
    • team onboarding
  2. 02

    AI products and agents

    Agents with tool access, search over company documents, e-mail and PDF processing, assistants on top of internal systems. Every feature ships with evals, logs, and a clear line where a human decides.

    • LLM APIs
    • tool use
    • RAG
    • evals
    • observability
  3. 03

    Systems that can carry it

    The model is only part of the solution. The rest is backend, CRM and ERP integration, APIs, permissions, security and tests. I've done that for years, and nothing reaches production without it.

    • .NET
    • TypeScript
    • integrations
    • E2E tests

02

Work

Anonymised at clients' request. Real problems, measured results.

01 · Mortgage brokerage
  • .NET 8
  • Claude API
  • PostgreSQL
  • Playwright

Mortgage paperwork in minutes, not hours

Problem

Advisors retyped data from bank statements, tax returns and employment contracts into an internal system by hand. One application took 2–3 hours, and errors only surfaced at the bank.

Solution

A pipeline that extracts structured data from uploaded documents, cross-checks them, and flags inconsistencies. The advisor only reviews and confirms. Plugged straight into the existing CRM.

−70%
time to prepare an application
96%
fields needing no manual fix
6 weeks
from discovery to production
02 · E-commerce
  • TypeScript
  • Next.js
  • OpenAI API
  • MCP

Customer support that knows where the order is

Problem

The support team answered hundreds of e-mails a day. Most were about order status and returns, but every reply meant clicking through three systems.

Solution

An agent that triages the e-mail, looks the order up in the warehouse and with the carrier, and drafts a reply. Simple cases go out in one click after approval; complex ones reach a human with full context.

58%
of e-mails handled in one click
4 h → 25 min
average response time
0
replies sent without approval
03 · Manufacturing
  • .NET 8
  • RAG
  • Elasticsearch
  • evals

An assistant over 20 years of technical docs

Problem

Knowledge about the machines was scattered across PDF manuals, service records, and the heads of two technicians close to retirement.

Solution

Search over documentation and service history with source citations. A technician on the shop floor asks from a tablet and gets an answer linked to the exact manual page. We built the eval set together with the technicians.

12,000
indexed documents
91%
correct answers in evals
−35%
time per service call

03

Engagements

Three ways to start. Always with a clear scope and price.

1–2 days

AI process audit

from €1,200

I walk through one specific process with your team and tell you what to automate, what not to, and what it will cost.

  • —on-site workshop
  • —review of data and systems
  • —written recommendation with ROI estimate
Book an audit

4–8 weeks

Pilot to production

from €9,900

One solution taken all the way to production on your data. Evals, monitoring and team training included.

  • —prototype within 2 weeks
  • —eval set and quality metrics
  • —deployment and integration into your systems
Discuss a pilot

ongoing

Embedded engineer

from 2 days a week

I become part of your team. I build further AI features, run what's in production, and support your in-house developers.

  • —regular iterations
  • —operations and security updates
  • —mentoring your team
Set up a retainer

Prices are indicative and exclude VAT. You'll get an exact quote after a free intro call.

04

How I work

Short cycles, measurable quality, relationships that last years.

  1. step 01

    People first, model second

    I walk through the process with the people who do it. I care about the data, the exceptions, and where today's solution breaks. Often half the problem is solved by plain software.

  2. step 02

    Prototype on your data

    In days, not months. On real documents and cases, so it's immediately clear what works and what's just a nice demo.

  3. step 03

    Evals before features

    A set of test cases we measure quality against. A prompt or model change is then verified, not guessed.

  4. step 04

    Operations and long-term care

    Handover isn't the end. Monitoring, updates, security patches, and further iterations based on how the system is actually used.

05

Experience

Fintech, startups, large platforms. Always close to the business.

public/pavel.jpgPD
  • 7+ years

    CTO — Premium Financial Services

    Responsible for all development at one of the leading mortgage brokers on the Czech market.

  • 2020–2021

    Senior Frontend Developer — Carvago

    Design and development of a large BusinessCase module for a pan-European platform.

  • startup

    Senior Frontend Developer — Vistag · UP21

    A product from Karel Janeček's accelerator. Among the best Czech startups at Startup World Cup & Summit, winner of the Fundlift Award.

  • startup

    Developer — Omnichannel platform

    Communication platform for a leading Czech telecom operator.

  • early days

    Websites and design — MediaEnergy s.r.o.

    A few hundred websites, mostly on WordPress. Graphic design too, which still pays off in product work.

  • education

    BSc — Global Business and International Management

    I studied management to understand what companies actually need from software.

06

References

What people I've worked with say.

  • “Pavel didn't show up with a ready-made solution, he showed up with questions. After a week he understood our process better than half the company, and two months later we had something in production we'd been talking about for a year.”
    Martin K. — COO, financial services
  • “Finally someone who knows both AI and plain old backend. No demo for management, but a system our people use every day.”
    Lucie H. — Head of Customer Care, e-commerce
  • “I appreciate that he told us openly where AI doesn't make sense. It saved us money and disappointment.”
    Tomáš V. — CEO, manufacturing company

07

Stack

I pick tools by the problem. These are the ones I know well.

AI

  • LLM APIs (Claude, OpenAI)
  • agents, tool use, MCP
  • RAG and search
  • evals and monitoring

Backend

  • .NET 8
  • CQRS, MediatR
  • DDD, TDD
  • PostgreSQL, MySQL, Elasticsearch

Frontend

  • TypeScript
  • React
  • Next.js

Testing

  • Playwright
  • E2E and regression suites
  • manual QA

Also

  • Node.js
  • PHP
  • Vue, Angular, Svelte

3D and games

  • Unreal Engine 5
  • Blender
  • own games since 2023

08 — Contact

Got a process a machine should be handling?

Send a few sentences about what's holding you back. We'll set up a short call and I'll tell you honestly whether AI makes sense for it.

pavel@dietrichdev.cz
ElsewhereLinkedIn ↗GitHub ↗