Ilshat Sharapov

AI & automation engineer

I ship AI agents, browser automations, and data pipelines that do real work — not demos. Below is what I've built recently. Most of it is running in production for an audience of one: me.

Case studies · 8

Philippines Living Map

Live

Interactive Philippines map — 200 curated locations, 9 hazard layers (earthquakes, typhoons, eruptions, floods, and more), year-based filters, lifestyle presets, dark/light theme, RU/EN. Live embedded below.

Leaflet.js Vanilla JS GeoJSON OpenStreetMap Overpass

Result: 200 locations × 9 hazard layers × lifestyle preset scoring. Fully static SPA — no backend, no build step, instant load.

Cebuano Learning Dashboard

Live

Personal language-learning dashboard built around a combinatorial method — 5 new words per day, auto-generated sentence permutations, spaced review. Live embedded below.

HTML CSS Vanilla JS LocalStorage

Result: Self-contained single-file app. Learning Cebuano (Bisaya) with a repeatable daily ritual instead of ad-hoc flashcards.

ThreadsAgent

Live

AI agent for Meta Threads that scans viral competitor content, generates original posts in a trained voice, and publishes via real-browser automation. Two live accounts (AI-tech + Dating niches).

Python Patchright (browser automation) SQLite Claude API LLM voice tuning

Result: Autonomous posting pipeline with 5 content layers (hot takes, observations, honest struggles, outcomes, questions). Real posts, real engagement, no Meta API.

AI Consulting Website

Live

Astro-powered landing site for AI consulting — designed to convert YouTube traffic into discovery calls. Built, deployed, iterating on copy.

Astro TypeScript Tailwind CSS Cloudflare Pages

Result: Production site on Cloudflare Pages with custom domain. Methodology-first positioning (data audits, AI workflows) over industry niche.

Blade & Crown — Barbershop Demo Site

Live

Demo marketing site for a premium barbershop — full-page editorial design with 18 AI-generated photos in a unified style. Shipped end-to-end including the content pipeline.

HTML CSS JavaScript Gemini API (image gen)

Result: Full marketing site (hero, services, gallery, before/after, CTA) with 18 consistent-style images generated via Gemini API. Live embedded below.

Whisper Typing

Live

Windows speech-to-text fork with enhancements — local faster-whisper, natural typing simulation, global hotkey, media pause, system tray. No cloud, no account.

Python faster-whisper PyAutoGUI Windows APIs

Result: Daily driver dictation tool. Forked upstream + added audio overlay, media pause, hold-to-record mode. Public fork on GitHub.

VideoAgent

In progress

Unified YouTube pipeline — one recording in, one long-form video + multiple shorts out. EDL-driven cuts, Whisper transcripts, MediaPipe face tracking, Remotion composition, FFmpeg render.

Python Whisper MediaPipe Remotion FFmpeg

Result: End-to-end EDL → long + shorts pipeline. 6/6 E2E passing, 94/94 unit tests green as of Day 6 build.

DeepResearch System

Live

A personal research system — 3 skills, 48 stored reports, 285+ sources. Structured pipeline for turning vague questions into citable, reusable knowledge. Live command-center embedded below.

Markdown YAML frontmatter Claude Code skills WebSearch + MCP

Result: 48 indexed research reports spanning AI tooling, content strategy, and technical deep-dives. Every claim is source-backed — every report is reusable context for future work.