Moutcome
A place for people and their ideas.
Architectural composition · not product UI AI Product Developer
I build
digital products.
With AI.
I design architecture, orchestrate AI agents, and turn ideas into working full-stack products.
Work Idea → architecture → validated productIdea → working product
DirectionProduct thinking
AssemblyAI orchestration
Drag to rotate
01 / Portfolio
Selected work06
Products, systems, and research. Different challenges. One considered approach.
MoutMessenger
Conversations. In real time.
WEB
DESKTOP
REALTIMESocket.io
03 / Custom AI ecosystemExperimental AI / R&D project
Mout AI
Not just using AI. Understanding how it works.
01RUN A MODEL
mAI Inference
A model inside a product.
Built a layer that makes an AI model usable inside an application.
Inference infrastructure02ADAPT A MODEL
mAI Sage
An existing model. Own tasks.
Adapted a Qwen-based language model for my own tasks.
Qwen · Fine-tuning · Model adaptation03BUILD A MODEL
mAI CPU
Own model. From the beginning.
Developed a custom language model of approximately 300 million parameters from scratch.
~300M parameters · Built from scratchLUMERINE
Foundation first. Platform next.
INFRASTRUCTURE01
SHARED CONTRACTS02
API + WEB03
NEXT / STAGE 0
Mout CPU Research
Measure. Validate. Decide.
A4
G128
G128
93.5212%Q4 gap closed · 4.1250 ATTENTION_Q BPW
Runtime: 42.83% / 43.85% of Q4 (n128 / n1024). Production winner: none.
More things I've built
Umber VPN
Commercial project · SoldA VPN service managed through a Telegram bot, subscription flow, and dedicated server infrastructure.
Designed for simple VPN access without an overloaded interface. The concept included automatic location selection, multiple devices per subscription, referrals, and subscription freeze functionality.
Detailing PWA
Commercial project · SoldA reusable PWA product for detailing centers and automotive service businesses.
A flexible foundation adapted through branding, services, media, contacts, and service-specific configuration.
Customers can choose a service and time slot, book without registration, and manage reservations. An owner dashboard provides analytics, booking management, and service administration.
React · TypeScript · Supabase · PWA
Telegram Parser
Working projectTelegram monitoring: collecting messages from selected chats and channels and filtering for a defined topic.
Instead of manual review, the system collects, analyzes, and filters the message stream. Sources and filtering can be adapted to each use case.
Umber CRM Bot
ProjectA Telegram-based tool for automating CRM workflows and customer interactions.
A compact operational tool moves everyday business processes into Telegram and reduces repetitive manual work.
02 / ApproachHow I build
How I build
From intention to assembly.
Each agent receives a bounded task, acceptance criteria, and independent verification. Product direction and final decisions stay human-owned.
Idea & product logicFrom idea to validated product
ArchitectureDevelopmentQA & review
Idea & product logic
Define the problem, the user, and the scope of a first working product.
Architecture & context
Design the system. Record constraints, specifications, and acceptance criteria in AGENTS.md, CLAUDE.md, and Skills.
Orchestration & build
Break work into tasks for specialized agents. Implementation moves in small, verifiable iterations.
Review & testing
Run independent review, QA, and security passes. Validate browser behavior and engineering quality gates.
Product & next iteration
Assemble verified parts, deploy the product, and return to the next decision. Direction and final judgment stay with me.
AI accelerates the work. The decisions stay with me.
AI is not a replacement for engineering judgment. It is an environment where product logic, context, and independent validation turn difficult ideas into working products faster.
Product ArchitectureAI Agent OrchestrationAI Development InfrastructureFull-stack Product DeliveryReview & QualityAutomation & IntegrationsResearch & Experimentation
Not one “best AI”.
A team for the task.
I design both the product and the process that builds it. Different models and specialized agents receive appropriate roles, shared context, and verifiable tasks.
Human-owned directionIntention. Boundaries.
Final judgment.
- 01Context & specification
- 02Tasks & acceptance criteria
- 03Independent review & QA
- 04Verified assembly → iteration
Different strengths. One intentional process.
These are typical roles, not fixed model assignments. The team adapts to the context; tool selection and assessment remain my responsibility.
ChatGPT
Product & strategy
Explore the idea, clarify the user journey, and define what a successful result looks like.
Claude
Architecture & independent review
Explore complex context, compare approaches, and critically review an engineering decision.
Codex
Implementation & validation
Work inside the codebase: focused engineering slices, tests, debugging, and browser verification.
Gemini
Research & a second perspective
Compare information and bring an alternative perspective before the final decision.
DeepSeek API
Integrations & experiments
Use the model through an API in practical integrations, prototypes, and automation.
Agents propose and implement. Independent validation catches mistakes. I decide what becomes part of the product.