# Alex Amyotov — AI Researcher, Consultant & Entrepreneur

> Alex researches how AI systems and large language models work in real-world conditions — how they handle context, how their behavior can be monitored and interpreted, and how to make their outputs more reliable.

Canonical site: https://amyote.com

## Bio

Alex researches how AI systems and large language models work in real-world conditions — how they handle context, how their behavior can be monitored and interpreted, and how to make their outputs more reliable. He builds projects at the intersection of art, media, and technology, and consults companies on implementing and optimizing AI pipelines.

For the past 25 years, Alex has worked in media and technology. He started as a reporter for Forbes and Newsweek, then founded Look At Media in 2008 — a publishing house behind Hopes and Fears, The Village, and Wonderzine. As CEO, he grew the network to over 30 million people across five countries — projects that helped define a generation of entrepreneurs and creators. The media brands were exited in 2020. In parallel, he co-founded Setka — a U.S.-based platform for structured content creation — acquired by TinyMCE in 2021.

Since 2015, he has been a committed believer in decentralization: blockchain, open-source infrastructure, and distributed systems as a fundamentally different way of organizing things.

He lectures at Bard College Berlin and collaborates as an expert and trainer with Thomson Reuters on AI in journalism and media.

If you want to discuss a project, collaboration, or just get advice — let's get in touch.

Full bio: [/bio.md](https://amyote.com/bio.md)

## Research

### Coke Ex Machina (2026)

*Narrative Agent · Experimental · Research · Simulation*

An experimental agent that operates a vending machine — and reads business books. The agent accumulates knowledge from management literature at the context and memory level (no fine-tuning, no post-training), then applies it to operational decisions in real time. An edge experiment in post-training knowledge integration and grounded agent behavior.

Markdown: https://amyote.com/p/r1.md · Page: https://amyote.com/#r1

### Mediacontext (2025)

*Context · LLM · Tool*

A research lab studying how AI reshapes media, propaganda, and information ecosystems — with a particular focus on regional and cultural gaps in large language models. Mediacontext investigates how LLMs encode bias, distort facts, and diverge across geographies, and builds open tools to surface and address these gaps.

Markdown: https://amyote.com/p/o6.md · Page: https://amyote.com/#o6 · Link: https://mediacontext.org/

## Consulting

### ElGranto — AI-First Grant Agency (2026)

*AI Transformation · Grant Writing · Workflow Automation · AI-First Org*

Region: Global • Industry: Grant consulting. Transformed a grant agency — helping clients write applications and secure funding — into an AI-first organization. Redesigned the full workflow around AI: from brief intake and document analysis to draft generation and compliance checks. Grant preparation time dropped from 2–4 days to 1 day, and the need to hire additional staff was eliminated entirely.

Markdown: https://amyote.com/p/c13.md · Page: https://amyote.com/#c13

### Support Service Assistance System (2025)

*Fraud Detection · Support · Marketplace*

Region: USA • Industry: Travel marketplace. Deployed an AI module that analyzes user correspondence and flags likely fraud patterns to reduce support load. In the first stage of rollout, support volume dropped by 30%. Validated team readiness for full operational deployment.

Markdown: https://amyote.com/p/c7.md · Page: https://amyote.com/#c7

### AI Training and Competency Verification for Coffee Chain (2025)

*HoReCa · Staff Training · Dialog Simulation · Dubai*

Region: Dubai • Industry: HoReCa. Built an interactive AI training platform for baristas and waiters: dialog simulations, menu knowledge checks, service standards validation, and progress tracking. Onboarding time cut to 2 days. Deployed across all staff and now moving to stage 2 — predictive hiring recommendations and adaptive training programs.

Markdown: https://amyote.com/p/c10.md · Page: https://amyote.com/#c10

### Monitoring Legal News and Legislative Changes (2025)

*Legal Ops · Monitoring · Summaries*

Region: CIS, MENA • Industry: Marketplace. Created an AI system (agent + Telegram bot) that aggregates legal news and regulatory/judicial updates across three countries, prioritizes by importance, and delivers short briefs directly to lawyers. Saves up to an hour per day previously spent scanning news feeds and professional communities — without missing critical signals.

Markdown: https://amyote.com/p/c11.md · Page: https://amyote.com/#c11

### Hyper-Targeted Ads Creative Production Pipeline (2025)

*Ad Production · Meta · YouTube · TikTok · Streaming*

Region: Global • Industry: Streaming (NDA). Built a production pipeline for hyper-targeted advertising creatives — 360+ assets per campaign, segmented by socio-demographic profiles across Meta, YouTube, and TikTok. Included automated variation generation, QA, and campaign-ready packaging for a major streaming platform and its vertical-series product.

Markdown: https://amyote.com/p/c12.md · Page: https://amyote.com/#c12

### Communication Tone and Style Control System (2024)

*Tone of Voice · Poland · LLM Ops*

Region: Poland • Industry: Development. Built a tone-of-voice control layer for marketing and contractor communications using a hybrid stack (GPT-4.0 + local Bielik), adapted to Polish language and cultural norms. Trained two team members and established a repeatable production process. The share of texts accepted without revision doubled after rollout.

Markdown: https://amyote.com/p/c6.md · Page: https://amyote.com/#c6

### Taxi Operator Quality Assessment (2024)

*QA Analytics · Contact Center · Transportation*

Region: MENA • Industry: Transportation. Implemented a full-coverage chat and call monitoring system with mood tracking (start and end of each interaction), resolution speed measurement, and quality scoring. Produced an operator performance rating used for coaching and accountability. Covers all operators and is live in production.

Markdown: https://amyote.com/p/c8.md · Page: https://amyote.com/#c8

### AI Trainer for Sales Managers (2024)

*Sales Enablement · Simulation · Agency*

Region: Russia, EU • Industry: Advertising Agency. Built a custom AI trainer for practicing sales calls and client messaging, tailored to the agency's context, key accounts, and stakeholder profiles. Trained the full sales force — six managers — bringing the entire team up to speed before live client conversations.

Markdown: https://amyote.com/p/c9.md · Page: https://amyote.com/#c9

## Open Tools

### PIPI-Claw (2026)

*Raspberry Pi · Security · Budget Optimization · NanoClaw Fork*

A fork of NanoClaw adapted for Raspberry Pi 4 and Pi 5. Focuses on two things the original doesn't prioritize: security hardening and token/API budget optimization. Started as a personal tool for home experiments and pet projects — now moving into client deployments as a lightweight agent runtime in production pipelines.

Markdown: https://amyote.com/p/o9.md · Page: https://amyote.com/#o9

### CTRL-L (2025)

*Terminology · Multilingual · Risk Notes · CMS · Guardrails*

A research project that developed a methodology for steering LLM behavior through structured glossaries. The effect was first discovered during translation pipeline design — and turned out to be far more broadly applicable. Glossaries enforce tone of voice, ensure content alignment, and counteract dataset contamination and training-time grooming. Validated empirically with editorial teams. Currently supports English, Russian, and Belarusian.

Markdown: https://amyote.com/p/o7.md · Page: https://amyote.com/#o7 · Link: https://huggingface.co/mediacontext

## Art & Fun

### Will it run? (2025)

*System Matcher · LLMs · OpenAI GPTs · Playful Utility*

A quick GPT tool for anyone curious about running LLMs locally. Describe your computer — RAM, GPU, chip — and get a practical list of which models will actually run on your hardware. No benchmarks, no marketing: just a straight answer. Platform: OpenAI GPTs.

Markdown: https://amyote.com/p/a1.md · Page: https://amyote.com/#a1 · Link: https://chatgpt.com/g/g-68f04690cc88819188f65519688e480e-will-it-run

### Conspiracy stays secret (2025)

*Science Explainer · Probability · Grimes Model · OpenAI GPTs*

Estimates how long conspiracies can stay secret using David Robert Grimes’ 2016 “conspiracy leak” model — in a fun, science-based way. Built as an entertainment tool, it doubles as a live demo in journalism training: Alex uses it in workshops at Bard College Berlin and with Thomson Reuters to show how AI can make fact-checking and probability reasoning genuinely engaging. Because good training should never feel like training. Platform: OpenAI GPTs.

Markdown: https://amyote.com/p/a2.md · Page: https://amyote.com/#a2

### Sailing Weather AI (2024)

*Sailing Forecast · Skipper Briefing · Weather · OpenAI GPTs*

Built for personal use — and for anyone who actually sails. A professional-grade weather briefer that speaks the skipper's language: wind in knots, gust patterns, wave height, sea state, and clear go / no-go guidance for safe navigation. Not a generic forecast — a decision-support tool. Tested across the Mediterranean and Adriatic. Platform: OpenAI GPTs.

Markdown: https://amyote.com/p/a3.md · Page: https://amyote.com/#a3