# AI & Full-Stack Development

Marketplaces, B2B SaaS, AI agents, internal tools — scoped, built, and shipped end to end by one person who has done it before. You don't need a technical background: that's what you're hiring.

## Overview

Key areas:
- AI agents & software factories
- Marketplaces & two-sided platforms
- B2B SaaS products you can sell
- Internal tools for non-technical teams

## Solutions

### AI Agents & Workflows

Agents that do real work instead of demoing well. They call your tools, read your documents, and get measured on whether the output is actually right.

- Agents with real tool access
- Answers grounded in your own data
- Evals so quality is measured, not hoped
- Failover across model providers

### Marketplaces & Platforms

Two- and three-sided platforms where supply, demand, and money all have to work at once. You bring the industry knowledge; I build the machinery around it.

- Listings, search & booking flows
- Payments, payouts & refunds
- Reviews, roles & verification
- Admin tools to run the operation

### B2B SaaS Products

A product your customers install and pay for every month. Multi-tenant from day one, so customer number two doesn't mean a rewrite.

- Multi-tenant accounts & billing
- Usage metering & dashboards
- Self-serve signup & onboarding
- API and SDK for your customers

### Internal Tools & Team Automation

Systems that let a non-technical team run technical work — brief a task, launch an experiment, approve a result, without anyone opening a terminal.

- Task, brief & approval workflows
- Control panels non-developers can use
- Scheduled jobs, reports & alerts
- Slack & Telegram as the interface

### Infrastructure & Code Audits

You already have something running and a nagging feeling about it. I go through the code and infrastructure and tell you what's fragile, what it's costing you, and what to fix first.

- Architecture & cloud cost review
- Security & reliability gaps
- Where the team is actually slowed down
- Ranked fix list with real estimates

### Idea Validation & Consulting

Sometimes you don't need a build yet. You need someone technical to pressure-test the idea, put a number on it, and say what to do first.

- Feasibility & honest cost estimate
- Build vs. buy vs. don't
- Stack choice & hiring advice
- A plan you can act on immediately

## Fixed-scope offers

Scope, price, and turnaround settled before the work starts.

### Codebase Audit

I read your whole codebase and hand back a prioritised list of what is broken, what it will cost you, and what I would fix first.

- Deliverable: Written report, prioritised
- Turnaround: 1 week
- Price: $960 (16 h × $60/h)

Includes:

- Security, performance, and architecture review
- The AI and domain-specific parts most reviewers skip
- Hosting and running costs, with the obvious savings
- A call to walk through the findings

### Website AI-Readiness Audit

I check whether ChatGPT, Claude, and search agents can actually read your site and act on it, and rank what to fix first.

- Deliverable: Report with a ranked fix list
- Turnaround: 3 days
- Price: $360 (6 h × $60/h)

Includes:

- `llms.txt`, markdown export, and crawlability
- Structured data and how agents parse your pages
- Agent-callable actions you are missing
- Fixes ranked by effort against impact

### 3-Day Proof of Concept

Your idea, clickable and running on real data in three working days, so you can show it before committing a budget.

- Deliverable: Deployed prototype
- Turnaround: 3 working days
- Price: from $1,080 (24 h × $45/h)

Includes:

- Working prototype on a live URL
- Real data, not screenshots
- Your repo and your accounts
- A call to decide what happens next

### Codebase Audit, Focused

The same audit, scoped to the one area you already suspect: security, performance, or the LLM layer.

- Deliverable: Written report, one area
- Turnaround: 3 days
- Price: $480 (8 h × $60/h)

Includes:

- One area, picked before we start
- Findings ranked by what they cost you
- Concrete fixes, not general advice

### Custom Software Factory

A pipeline of agents that runs one job your team repeats every week — the same migration, the same integration, the same ticket-to-PR pass — with the checks that make its output worth trusting.

- Deliverable: A running pipeline, handed over
- Turnaround: 3 weeks
- Price: $2,700 (60 h × $45/h)

Includes:

- One job, picked because it repeats
- Agents wired to your repo, your tools, and your data
- Quality gates: evals, tests, and a human step where it matters
- Handover, so your team runs it without me

### AI-Readiness Implementation

The AI-readiness fixes shipped to production, so agents can read your site instead of guessing at it.

- Deliverable: Merged pull requests, live
- Turnaround: 2 weeks
- Price: $900 (20 h × $45/h)

Includes:

- Markdown API for every page
- `llms.txt` and structured data
- Agent-callable endpoints
- Verified against real agent traffic

### Technical Due Diligence

For a founder or investor reviewing a build someone else delivered: whether the code is worth what you paid and what it costs to keep.

- Deliverable: Diligence report, plain language
- Turnaround: 1 week
- Price: $960 (16 h × $60/h)

Includes:

- What you own and what you are locked into
- Cost to run and cost to maintain
- Whether a new team can pick it up
- Written for a non-technical reader

### Idea Validation

Ninety minutes and a written plan covering what to build first, what it costs, and whether to build it at all.

- Deliverable: Build plan, one page
- Turnaround: 2 days
- Price: $200 (2 h × $100/h)

Includes:

- 90-minute call on your idea
- Scope for the first version
- Cost and timeline you can budget against
- Credited against a build if you go ahead

### Search Upgrade

Your keyword search replaced with hybrid vector search and reranking on one dataset, measured before and after.

- Deliverable: Search shipped, with a benchmark
- Turnaround: 2 weeks
- Price: $1,440 (32 h × $45/h)

Includes:

- Embeddings and hybrid retrieval on one dataset
- Reranking tuned to your queries
- Relevance benchmark, before and after
- Runs in your infrastructure

### 2-Week MVP

A live product in front of users in two weeks, with auth, payments, and analytics wired up.

- Deliverable: Live product, taking users
- Turnaround: 2 weeks
- Price: from $3,600 (80 h × $45/h)

Includes:

- Auth, payments, and analytics
- Deployed and instrumented
- Your repo, your accounts, your code
- Handover, or keep me on to iterate

### Fractional AI Engineer

A standing slot in my week to ship features, review your team's AI-generated code, and keep the stack current.

- Deliverable: A reserved slot each week
- Turnaround: Starts within a week
- Price: $1,800/mo (40 h × $45/h per month)

Includes:

- Features shipped against your roadmap
- Review on AI-generated pull requests
- Async access between sessions
- Cancel any month

## Reliability

### It doesn't fall over at 3am

Every failure is handled on purpose rather than discovered in production. Pickart's live backend runs on Effect-TS, where errors are values the code has to deal with, not surprises.

### One change can't silently break another

One type system runs from the database to the screen, validated at every boundary. Change a field and everything that depends on it breaks at build time, in front of me — not later, in front of your customer.

### Tested before it reaches you

Automated tests click through the real product on every change, and the deploy is blocked if anything fails. No manual "hope it still works" pass.

### AI features survive a bad day at OpenAI

Model providers do go down. Yours switches to a backup automatically, which is how Outrizz stayed up through outages that took other products offline.

### You're not locked into my choices

Your business logic sits apart from the tools underneath it, so swapping a database, a payment provider, or a model later is a change — not a rebuild. Recruit AI is built this way.

### You can see what's actually happening

Logs, traces, and error alerts are wired in from the first deploy. When something looks wrong we know which request, which user, and which line — instead of asking you to reproduce it.

## Featured Work

- https://andrey-markin.com/projects/pickart
- https://andrey-markin.com/projects/outrizz-ai
- https://andrey-markin.com/projects/recruit-ai
- https://andrey-markin.com/projects/aggregator
- https://andrey-markin.com/projects/ai-form
- https://andrey-markin.com/projects/telegram-claude

## Tech Stack

React, Next.js, TypeScript, Node.js, Go, Python, OpenAI, Anthropic, PostgreSQL, Supabase, Drizzle, Stripe, Expo, Telegram, Slack, Vercel

## FAQ

### I'm not technical. Can we still work together?

That's most of my clients. You know your customers and your market; the technical decisions are my job, and I'll explain the ones that affect your money or your timeline in plain language. You'll never be asked to choose a database.

### I have an idea, not a spec. Is that enough to start?

Yes. We start with one call where I ask questions until the idea becomes a concrete plan: what it does, who it's for, what the first version has to include, and what it will cost. You get that plan whether or not you hire me for the build.

### Can I just buy advice instead of a build?

Often that's the right call. I do idea validation, technical due diligence, and infrastructure audits as standalone work — a few hours or a fixed-fee review, with a written plan at the end. If the answer is "don't build this yet", I'll say so.

### If you build fast, can I trust the quality?

Speed comes from the foundations, not from cutting corners. Errors are handled by design, the whole stack shares one type system with validation at every boundary, automated tests run before anything deploys, and monitoring surfaces problems early. The same code serves real, paying users today.

### Who owns the code and IP?

You do. Everything I build is yours — full source, in your repository, on your accounts, with no lock-in to me. I hand it over so your team or your next developer can pick it up cleanly.

### What's a typical timeline?

A working prototype in 3–5 days, an MVP you can put in front of users in about 2 weeks, then we iterate on real feedback. Larger platforms grow from there in short, visible increments — you always have something to look at.

### How does pricing work?

Hourly or fixed per project — see pricing below. Small, well-scoped work is usually fixed-price; longer or evolving builds work better hourly. I'll recommend whichever costs you less for what you actually want.

### What if the prototype shows the idea doesn't work?

Then you've risked a few days instead of a full build — which is exactly why we start there. Finding out early is a good outcome, and it means your budget goes to the version that does work.

### What stack do you build on?

TypeScript end to end. Next.js, Vite + React, or TanStack Start on the front end, React Native and Expo for mobile. Node or Go on the back. PostgreSQL via Supabase, PlanetScale, or Convex, with Drizzle. Stripe for payments. The Vercel AI SDK keeps the AI layer provider-agnostic, so models can be swapped and output can be measured. Right tool per project, not one stack forced onto everything.

## Contact

- Website: https://andrey-markin.com/services/full-stack-ai-development
- Contact form: https://andrey-markin.com/#contact