Štefan Bohumel - AI Automation | ContraWork by Štefan Bohumel
Štefan Bohumel

Štefan Bohumel

Python AI Dev | ComfyUI, LoRA/LLM Training, Web3, Web API

    $1k+
    Earned
    18
    Followers
Followed by Sian V, Sharon O, and Sharyar K

Olomouc, Czechia

WorkPostsAbout

Olomouc, Czechia

Contra x Lovable Challenge Submission: Monolith Tattoo Atelier 1. Business Archetype & Profile Business Name: Monolith Tattoo Atelier Location: Berlin (Kreuzberg), Germany Solo Operator: Alex Vance (Specializing in fine-line, blackwork, and sacred geometry) Public Application Link: https://monolith-tattoo-atelier.lovable.app 2. Client Brief & Operational Quirks Alex is a solo tattoo artist operating at peak capacity, spending 6 to 8 hours daily wearing sterile gloves with a running machine. While working, he is physically and hygienically incapable of answering calls or responding to messages. Prior to this solution, his business suffered from three critical bottlenecks: Instagram DM Chaos: Receiving 40+ inbound inquiries daily with vague parameters ("How much for something medium on my arm?"), creating 15+ hours of unpaid administrative back-and-forth weekly. Unscoped Appointments: Clients cannot accurately specify dimensions in centimeters, assess placement difficulty, or identify skin contraindications (fresh scars, active eczema, moles). Chronic No-Shows: A 28% no-show rate caused by booking calendar slots without upfront financial commitment or medical pre-screening. 3. One-Line Description of the Problem Solved Turned vague, open-ended Instagram DMs into deposit-secured, anatomically-scoped tattoo bookings with zero manual back-and-forth from the artist. 4. Technical Architecture & Key Innovations Built in Lovable The application solves both the "Front Door" (client intake) and the "Follow-Through" (studio management) in a synchronized dual-view architecture: Interactive Anatomical Body Selector: Replaces vague text descriptions with a vector silhouette featuring front/back toggles, realistic contouring, and a dynamic Heatmap Pain Index that automatically flags high-sensitivity zones (ribs, neck, spine) requiring session buffers. In-Browser Stencil Composer: Solves the primary hesitation in custom tattoos. Clients can position, scale, rotate, and preview artwork directly over their own skin using authentic 'multiply' blend modes before stepping into the studio. Dynamic Session & Pricing Engine: Calculates exact session tiers (Quick, Half-Day, Full-Day) based on size (cm), anatomical placement difficulty, and artistic style, automatically locking compatible calendar blocks and deposit amounts. Guaranteed Escrow & Medical Gate: Enforces a 4-point safety declaration and a non-refundable Stripe deposit checkout, reducing no-shows from 28% to 0%. Artist iPad Back-Office & Day Sheet: A dedicated studio control dashboard displaying live synchronized appointments, escrow totals, and an iPad Day Sheet with high-res stencil specifications, medical clearances, and outstanding balances due upon arrival. Automated DM Deflection: An integrated router configuration that deflects inbound social inquiries straight into the scoped booking funnel. 5. Attached Deliverables Product Demo Video (< 3 minutes): [https://youtu.be/OmYEf88Qe_U ] Comprehensive end-to-end walkthrough demonstrating both the customer booking journey and the live artist back-office synchronization. Process Walkthrough Video (Bonus): [https://youtu.be/eiHLxgql1b4 ] Fast-paced engineering walkthrough showcasing the Spec-First development methodology, phased prompt injection, and code hardening in Lovable. Architectural Specification (Markdown): [https://github.com/Ra9mirez11/Monolith-Tattoo-Atelier/blob/01cae1362b10ef001ee2606415491dcc673707a3/Lovable%20plan%20atelier%20tattoo.md ] Initial domain blueprint, data schemas, mathematical formulas, and operational states drafted prior to code generation. Phased Prompt Engineering Log (Markdown): [https://github.com/Ra9mirez11/Monolith-Tattoo-Atelier/blob/main/Prompt%20for%20lovable%20.md ] Chronological record of all 12 systematic prompts utilized to construct and optimize the production build.
1
128
RECLAIM - Gamified Eco-Volunteering Working Prototype The Problem Traditional eco-volunteering often lacks engagement, making it difficult to motivate individuals, especially the younger generation, to participate consistently. The process of reporting and verifying cleaned areas is manual, inefficient, and untrusted. The Solution: RECLAIM RECLAIM transforms eco-volunteering into a real-life RPG, making healing the planet fun, competitive, and highly rewarding. Users create an account, secure it via Supabase email authentication, and select their unique 3D Eco-Spirit avatar. The core innovation is our automated verification loop: Users find cleanup missions on a real-time map. They complete the mission and take an "after" photo. We integrated Gemini AI to perform a strict direct comparison between the "before" task image and the new "after" photo. There are no bypasses—Gemini AI decides if the location is pristine. Upon AI verification, the Supabase backend awards XP and unlocks premium 3D badges. The Advanced Figma AI Workflow This project serves as a showcase for the advanced capabilities of Figma's new AI toolkit: • Figma Agent was used to generate and iterate upon the entire multi-screen design flow, strictly adhering to our premium 'Eco-Glassmorphism' style. • Figma Weave generated all procedural 3D visual assets, including Eco-Spirit companions and collectible badges. • Figma MCP was utilized to integrate live, data-driven satellite maps directly into the dashboards. • Figma Make exported the completed design into a functional Next.js/React application, making it a working prototype. Live Links Working App: https://path-os-41270623.figma.site Figma Community File: https://www.figma.com/community/file/1648087862770865252 (https://www.figma.com/community/file/1648087862770865252)Pitch Video: https://youtu.be/skytzF6qljY Technology Stack • Frontend: Next.js (exported via Figma Make), React, Tailwind CSS • Design: Figma (Agent, Weave, MCP) • Backend: Supabase (Auth & Database) • Verification: Gemini AI (Visual Comparison)
6
8
428
Cover image for While most participants naturally lean
While most participants naturally lean towards building standard web apps or generic startup dashboards—concepts the industry is already highly familiar with—I decided to bring the theme 'Interfaces that feel alive' to a space that is notoriously static, flat, and outdated: the professional resume. I built 'The Living Portfolio,' a next-generation interactive CV tailored for a hybrid Web3, AI, and Security Engineer. By breaking away from traditional dead PDFs, this interface uses premium dark glassmorphism, tactical micro-interactions, and responsive canvas layers to transform a career summary into a breathing, high-fidelity digital identity that actively engages the user. I utilized Google Stitch as the core engine for establishing the initial responsive layout and grid structure. From there, I transitioned into an iterative process using in-place edits to inject custom CSS motion properties, technical typography, and interactive button states. Finally, despite heavy platform limitations regarding multi-page structures, I successfully leveraged Stitch's export capabilities to deploy the compiled pages directly onto Netlify, achieving a fully functional and publicly accessible live prototype. While the core concept of Google Stitch is highly innovative and the visual potential of the canvas is impressive, the developer interaction with the AI agent was a deeply frustrating experience filled with friction. During the detailing phase, the agent consistently suffered from severe context drift. It repeatedly violated explicit constraints—such as generating entirely new layout windows instead of modifying existing components in-place as instructed. Furthermore, it frequently erased pre-existing text data and wiped out working elements during subsequent prompts. It felt as though the agent is programmed to over-generate new code overzealously rather than respecting precise, incremental boundaries. While I see the immense value in Stitch as a builder, managing the agent for high-fidelity UI work currently requires an exhausting amount of backtracking to fix forgotten or deleted elements. P.S. I'm honestly not 100% satisfied with the final result, but iterating with the agent is a never-ending loop—with every new edit, something else kept getting lost or forgotten :-D. So, I'm submitting it as is and going to chill. https://stitch.withgoogle.com/projects/4307425184075552527 https://stefan-bohumel-resume.netlify.app/ https://github.com/Ra9mirez11/stitch-projekt
1
5
402
Cover image for Mantle Agentic Treasury Manager (ATM) Autonomous
Mantle Agentic Treasury Manager (ATM) Autonomous AI-driven DeFi portfolio management on the Mantle Network Mantle ATM is a fully autonomous, on-chain AI treasury manager deployed on Mantle Sepolia. It leverages Llama 3.3 70B (via Groq and LangChain) to continuously analyze yield opportunities across the Mantle ecosystem — specifically mETH (Mantle Liquid Staking) and USDY (Ondo RWA yield) — and autonomously executes reallocation transactions on-chain to maximize yield while enforcing strict on-chain security guardrails. The Innovation: Circular Agentic Economy What makes Mantle ATM unique is its self-sustaining circular tokenomics loop: The Roast Chamber: Users pay a micro-fee of 2 MNT in an on-chain transaction. The Product: Llama 3.3 scans the user's wallet holdings and generates a brutally hilarious AI Roast (or glowing Toast) that they can immediately share on X (Twitter). Closing the Loop: The collected MNT fees are deposited directly into the treasury gas reservoir. The autonomous AI Agent uses these funds to pay gas for its scheduled hourly rebalancing crons, ensuring the agent runs forever without needing manual developer refills. Key Features Deployed Mantle Ecosystem Live Ticker A continuous, marquee-style scrolling ticker running directly beneath the navbar. Powered by a keyless CoinGecko Live API fetch (MNT, mETH, USDY, USDC) with local cached fail-safes. Premium Interactive Charting (Alpha Delta) Rich technical APY charts showing projections, horizontal gridlines, extrema markers, and teal area gradient fill under the curve. Features custom hover crosshairs and floating tooltipsdisplaying Time, APY, Volume, and Reward attribution. Includes a DeFi metrics dashboard (TVL, Utilization, Collateral Ratio, LP Summary, and Volatility). Dedicated Portfolio Hub (/portfolio) An isolated sub-route providing a deep On-Chain Wallet Scanner that lists connected wallet balances (USDC, USDY, mETH, MNT) on Mantle Sepolia. Features a Yield Allocation Optimizer advising on asset distributions. Smart Money Whale Scanner Tracking and interpreting major Mantle Network "whale" addresses (Treasuries, high yield stakers, arbitrage nodes) via Llama 3.3. 🔥 AI CFO Roast & Toast Chamber Gamified Web3 on-chain utility that lets users burn 2 MNT to unlock AI roasts of their wallet assets. Technical Stack & Architecture Frontend: Next.js 15, React 19, TypeScript, Tailwind CSS, Framer Motion Web3 Layer: RainbowKit, Wagmi, Viem (direct JSON-RPC node reads) Backend API: Flask (Python 3.10) hosted on Vercel Serverless Functions AI Brain: LangChain, Groq API, Llama 3.3 70b Model Smart Contracts: Solidity 0.8.x, compiled and deployed via Foundry Automation & Alerts: Vercel Cron Jobs, Discord Webhooks Strict On-Chain Security Guardrails The AI agent cannot move funds to arbitrary addresses. Our verified smart contracts enforce: Token Whitelist: Only native USDC, USDY, and mETH can be reallocated. Access Control: Only the designated AI Agent wallet can call reallocate(). Amount Cap: Max 500,000 USDC reallocated per autonomous cycle to prevent slippage. No Withdrawal: Funds stay in TreasuryManager contract at all times.
2
325