DANYLO PRAVDA

BUSINESS AUTOMATION ENGINEER — AI-AGENT SYSTEMS — RESPONDS WITHIN 24H

Software that runsthe business.Busywork, done for you.

I build AI-agent systems that take repetitive work off your team's plate — image production, monitoring, reporting, internal tools. If a process in your business runs on copy-paste and patience, it can probably be a system instead. (The same machinery built and runs a 728,000-line trading system, solo, in 83 days — so it holds up under load.)

M.01 — COMMITS SINCE GOING ALL-IN

0

Across 8 repositories in 83 days (2026-03-21 to 2026-06-12) — sole developer, about 40 commits (saved, documented units of work) a day — the throughput of a small team, from one person.

M.02 — DAYS FROM PIVOT TO THIS PORTFOLIO

0

First 2026 commit (btc-bot, 2026-03-21) to the latest commit (2026-06-12)

M.03 — LINES OF CODE AND DOCUMENTATION

0k+

Total text lines across 11 portfolio repositories; forked third-party code excluded

M.04 — PROJECTS DOCUMENTED

0

From a 728k-line trading system to client tools shipped in one day; includes pre-2026 production tooling

№ 01 — FLAGSHIP

BTC 5-min Polymarket Bot

PROBLEM

Trading Bitcoin's 5-minute prediction markets means reacting in under a second across 11 live exchanges — normally a four-person team's job.

SOLUTION

One engineer built and runs it solo using AI agents: a production Rust system that reads 11 exchanges, scores an ML model, and prices each bet in microseconds.

OUTCOME

A live system on AWS (paper mode, no real money) that decides in 21 microseconds — locked behind 65 frozen design records, 13 unskippable quality gates, and 102 stress tests, the same discipline you get on simpler automation work.

291M / day
Market events captured per day
21 µs p50
Decision speed, in production
2,948 passing
Automated tests
13 strict gates
Quality gates per commit

This is what one engineer can manage when the agents do the typing.

 

 

Read the full case study

№ 02 — SELECTED WORK

P.01

AI Image Compositor for CRE Marketing

CLIENT WORK

409 Commits in 7 days

P.02

Autonomous Bug-Bounty Pipeline

AI SYSTEMS

4 Platforms monitored

P.03

Watch Catalogue Operations Hub

CLIENT WORK

~290 Renders delivered

P.04

Yacht Assets Reviewer

CLIENT WORK

115 Assets reviewed

P.05

Claude Code Kit

TOOLING

6,689 Source lines (Python)

P.06

Cinema 4D Render Takes Automation

AUTOMATION

1,988 Takes rendered unattended

ALSO IN THE CATALOGUE

P.07

Nodeger

TOOLING

88 Source files

№ 03 — THE SPECIFICATION

AI & Agent Systems▾
  • AI agent orchestration (Claude Code, unattended background sessions, supervisor processes)
  • Multi-agent swarm design with crash recovery and quota handling
  • LLM pipelines: vision analysis, Whisper transcription, AI judges
  • ML inference (ONNX Runtime, XGBoost, CatBoost)
  • Token and context optimization for LLM workflows
  • AI quality gates: code critics, pre-commit LLM judges
Business Process Automation▾
  • End-to-end workflow automation with crash-safe resume
  • 24/7 monitoring, change detection, and watchdog systems
  • Multi-channel alerting (Telegram, email, voice call, GitHub)
  • AI API orchestration with hard spending caps and audit logs on every paid call
  • Batch pipelines: rendering, image compositing, export handoff
Engineering at Scale▾
  • Rust systems programming (Tokio, lock-free shared-memory IPC)
  • Python backends (Flask, NetworkX, pytest)
  • Architecture Decision Records and enforced quality gates
  • Quantitative stress testing (soak suites) and live-vs-simulation parity checks
  • AWS deployment, ARM cross-compilation, automated rollback
Web & Client Tools▾
  • TypeScript / React (React Flow, Konva, D3.js)
  • Fast internal tools: review, approve, export workflows
  • API integrations (GitHub, Dropbox, platform APIs)
  • Serverless and cloud deployment (Vercel, Render, AWS)
Creative Production Pipelines▾
  • Blender and Cinema 4D render automation
  • Computer-vision compositing (OpenCV auto-align)
  • CSV-driven state machines for production tracking
  • 5 years of senior 3D design for international brands

№ 04 — SERVICES

START SMALL

Most engagements begin with a fixed-scope first build: one process, one working slice on your real data, delivered in days. Several tools on this page were scoped, built, and live in a single day. You see real output before committing to more.

S.01

Internal tools, delivered fast

The usual first project: a single purpose-built tool, scoped and live in days, so you see how I work before committing to anything larger. Purpose-built web tools for review, approval, and handoff workflows — replacing spreadsheets and email chains. Demonstrated: a multi-user asset review app scoped, built, and deployed in one day, and a conversational project tracker used on real production work.

S.02

Workflow automation with AI agents

Take a process your team does by hand — image production, data preparation, catalogue work — and turn it into a supervised pipeline with spend caps, audit logs, and crash-safe resume. Demonstrated: a compositor that compresses 2–3 days of manual image work into one supervised sitting, and a 240-variant catalogue run finished in under five minutes.

S.03

Always-on monitoring and response

Systems that watch sources around the clock, detect the changes humans miss, alert you on every channel you care about, and kick off automated work in response. Demonstrated: a pipeline that polls four platforms every 15 minutes, catches quietly-edited pages by comparing content fingerprints (hashes), so even unannounced changes are detected, and launches AI research workers on trigger.

S.04

Operations systems for recurring work

Documentation-first hubs plus scripting that make complex recurring jobs repeatable and auditable, so the second engagement costs a fraction of the first. Demonstrated: a catalogue operations hub for a Swiss luxury watchmaker (Geneva) covering 290 delivered renders across four collections.

№ 05 — PROCESS

01

Map the work

A short call and a screen recording of the process as your team does it today. We name what it costs in hours, what done looks like, and the measurable criteria the system has to pass.

02

Working slice in days

I run parallel AI agents to compress build time, so the first end-to-end run on your real data happens in days, not weeks. You see real output early and steer before anything is locked in — which means you judge whether it's working on your own data, early and cheaply, before committing to the full build.

03

Harden

Spend caps, audit logs, crash recovery, kill switches, and an operator review surface. The system is tested against its failure modes, not just the happy path. Your data and credentials stay in your environment — secrets are never committed to code, API spend is hard-capped, and every action is logged so you can audit exactly what ran.

04

Hand over or operate

Full documentation and a git history of every change. Your team runs it, or I operate it and keep improving it as the work compounds.

№ 06 — ABOUT

I spent five years as a senior 3D designer for international brands, always with scripts under the design work — including a production pipeline that automated the rendering of 6,000 CGI videos.

In March 2026 I went all-in on AI-driven development.

The numbers from 83 days of full-time work: 3,344 commits across 8 repositories, all solo, working through orchestrated AI agents — currently #1 worldwide on ccgather, the public leaderboard ranking the heaviest users of Claude (Anthropic's AI) by volume. The flagship is a sub-second trading system: 15 Rust modules, 2,871 commits, 65 locked Architecture Decision Records — every major design choice documented and frozen — and 102 automated stress tests that run against live data. One person, with the audit trail of a small engineering team.

The client work applies the same machinery to ordinary business problems. An AI image pipeline that compresses 2–3 days of manual marketing work into one supervised sitting. A watch-catalogue production system that composites 240 product variants in under five minutes. A review-and-approve tool scoped, built, and deployed in a single day.

I am early in this business — about two months of focused automation work for clients. What I can show is speed, scale, and discipline: every number on this page comes from a git history you can audit. If your team repeats the same manual work every week, that is a system waiting to be built.

GITHUB
DPRVDA
EMAIL
dprvda@gmail.com
BASE
REMOTE
RESPONSE
WITHIN 24H

OBJECTIONS, ANSWERED

You're solo — what if you're unavailable?

Every system ships with full documentation and a complete git history of every change. Your team can run it, or hand it to anyone else — the watch-catalogue hub was built specifically so a successor could pick it up cold.

You're early in this business — why trust you with production work?

The track record is the audit trail: 3,344 commits across 8 repositories in 83 days, every number on this page traceable to a git history you can inspect. Behind the client work is a 728,000-line system carried solo with the discipline of a team.

What if the automation breaks?

Systems are built with spend caps, kill switches, crash recovery, and an operator review surface — and tested against their failure modes, not just the happy path.

ELEVEN SYSTEMS — ONE ENGINEER — ONE POINT OF CONTACT

№ 07 — START A PROJECT

Write to me.

Tell me which repetitive process eats your team's hours — I'll reply within 24 hours with whether it's automatable and what a first build would look like.

dprvda@gmail.com
GITHUB

Working with clients across the US and Europe (NYC, Geneva) — remote, overlapping both timezones. Currently taking on new projects.