AI Receptionist Call Workflows
DeployedPost-call automations for AI phone agents: when a voice agent finishes a call, the caller is captured as a lead and a call summary is sent, plus demo flows covering both sides of a sample call.
The systems portfolio
Everything below was designed and built by Scott Holmes and is real: running in production, in daily use, a working prototype or clearly marked as still in development. It is the reason we publish fixed prices instead of writing bespoke quotes. Most of your build is already proven somewhere on this page.
It's not our business on the line. It's yours. Scrutinize anyone you hire to build AI, and that includes us.
A live system doing real work, not slides or a scripted demo. Ask us to show you any system on this page running.
Talk to someone who has used the system for months, not days. Ask what broke and how fast it was fixed.
Every system on this page is labelled. If a vendor can't tell you which of theirs are in production, that's your answer.
Test results, audit logs, error rates, real timelines. "AI-powered" is a label, not evidence.
The workflows, the prompts, the accounts and the data. If you stop paying, does it keep running?
Where a person approves, what gets logged, and how a mistake gets caught before your customer sees it.
The build price and the monthly running cost, before any work starts. A quote that can only go up is not a price.
A working prototype on your own sample data beats any sales call. Ours is free, in 5 business days.
Ask us all of it. Then ask everyone else the same questions.
A bespoke quote prices your project as if nobody had ever built anything like it. We almost always have. Agents, approvals, audit logs, voice front desks, document extraction, CRM and accounting integrations, private retrieval over your own files: the pieces are already built, tested and running. AI coding tools then do the remaining build work about 5x faster. We pass both savings on, as published prices and a free working prototype in 5 business days.
2 systems
Post-call automations for AI phone agents: when a voice agent finishes a call, the caller is captured as a lead and a call summary is sent, plus demo flows covering both sides of a sample call.
An AI tutor a student can type or talk to and hear answers from. It asks short diagnostic questions to find the gap, explains at the depth asked for, answers only from a certified course knowledge base and an open textbook with a citation on every fact, and tracks what the student has mastered so it can run a short heads-up session before hard material.
3 systems
Scheduled Claude agents run the marketing operation: a weekly head-of-marketing review prioritises the hypothesis backlog and directs three daily specialists for content and social posting, website conversion health, and growth analytics, alongside daily certificate checks across every domain.
Sorts incoming email across several accounts into Important, Normal and Low, checks Sent Items so already-answered threads are not resurfaced, tracks soft commitments as open loops, and sends a six-hourly digest of what actually needs attention.
A monday.com marketplace app in build: ask monday's Sidekick assistant to log a receipt attached to a board item, and the tool fetches the file, has Claude extract vendor, amount, currency, date and category with per-field confidence, and creates a 'Pending Review' expense row with the receipt attached.
7 systems
Built for a Canadian consumer lender
Moves a lender's daily customer-records export from SharePoint to its card-payment processor over key-based SFTP, with error notifications and an independent watchdog that confirms the transfer happened, replacing a weekly manual upload. It runs on a self-contained n8n platform first delivered as a Docker package and now cloud-hosted.
Built for a Canadian consumer lender
Keeps the customer login portal on a lender's WordPress site current by uploading the loan system's account export automatically. It began as a headless-browser automation that drives the admin import screen and writes success or failure marker files, and was later replaced by a scheduled FTPS upload that the site's importer picks up.
Built for a Canadian consumer lender
Each morning sends a lender's customer screening batch to a sanctions and watch-list screening service: pulled from SharePoint, unzipped, PGP-encrypted by a small serverless microservice and uploaded by SFTP. Later it collects the results, decrypts them, flattens the XML into a readable CSV, files it in SharePoint and emails the analysts.
Watches any list of web pages on a schedule, extracts a signal (raw content, a price or a stock status), compares it with the previous snapshot and reports what changed, optionally posting each change to a webhook for Slack, email, n8n or another app.
A published n8n community node, built on the open-source community PGP node, that does one job: encrypt text or files with a recipient's PGP public key, so a workflow can, for example, encrypt a CSV before an SFTP upload. It removes private keys, passphrases, signing and decryption to cut failure points, and repairs common copy-paste damage to public keys before parsing them.
Built for a wireless and telecom engineering firm
Keeps a field team's WhatsApp work groups in sync with Monday.com: every message, photo, video, voice note and document is filed by group and by day on the right board, with attendance and issue tracking, a daily site report with a photo log, a manager dashboard and health alerts. Crews keep using WhatsApp exactly as they do today, and Claude can read the synced history to write reports.
Built for a wireless and telecom engineering firm
Every two weeks, queries a Monday.com board for change requests not yet billed, builds invoice rows in the ERP's import layout (customer, PO, due date, tax code, department) and emails finance a ready-to-import CSV.
12 systems
Given a domain or website, crawls the company's own site and returns a structured firmographic profile: name, description, emails, phone numbers, postal address, social profiles and a detected technology stack, for one company or thousands.
Cleans and scores email lists: syntax, MX records, disposable and free-provider domains, role addresses, an optional SMTP mailbox probe and catch-all detection, returning valid, risky or invalid plus a 0 to 100 deliverability score. It consumes any list or another scraper's dataset, so it slots under any lead-generation workflow.
Turns a niche and a city into a clean, de-duplicated, CRM-ready lead list: businesses come from a Google Maps source, then each website is crawled for the email, extra phone numbers and social profiles that Maps does not provide, with placeholder emails and bare platform links filtered out.
Pulls public procurement notices from UK Contracts Finder, US SAM.gov and any Open Contracting (OCDS) portal into one schema, filters by keyword and recency, removes notices already seen in earlier runs and can push new tenders to a webhook, so it runs as a scheduled bid-opportunity radar.
One webhook pipeline receives contact and lead forms from every site in the domain network, normalises and spam-checks each submission, stores it and alerts the owner within seconds.
A config-driven prospecting pipeline that scrapes local businesses by niche and city, enriches and verifies their emails, qualifies and profiles each lead, drafts a personalised first email with Claude and pushes it to a cold-email platform with A/B framing, per-domain throttling and suppression.
Scores a prospect's current website on six dimensions to decide whether a redesign pitch is worth it, then generates three genuinely different website samples (layout, look and strategy) and publishes them live for the prospect to react to.
Scrapes a provincial government tenders portal without logging in, flags the opportunities that fit by commodity code and keywords, and keeps a ledger of deadlines, status and next actions. Amendments between runs come out as a digest. Buyer documents for live bids are text-extracted and indexed so their requirements can be searched and compared.
Scans chosen subreddits for recent posts, scores each for relevance to the business's services, engagement and recency, flags the response window before a thread goes cold and suggests a value-first reply angle, with handled and dismissed tracking in the dashboard.
For any industry and city, pulls local businesses from Google Maps, scores them by real-world reputation, and checks whether ChatGPT, Perplexity and Google's AI actually recommend them. It surfaces the well-reviewed businesses AI overlooks (the sales opportunity for an AI visibility agency) and ranks every business as AI-and-customers-agree, AI-hidden winner or AI-overrated.
An operator dashboard that runs local-business lead discovery through n8n (search results, normalization, scoring), stores companies and opportunities in Postgres, and turns them into a working lead pipeline with notes, tasks, stages, duplicate detection and a do-not-contact list. The browser never sees raw n8n webhook URLs: the backend brokers every workflow start and resume.
A clean-room rebuild of Prysio. All AI moves to the server, and free users are scored on their own device so nothing leaves the browser. It adds a per-contact memory timeline with reply and outcome capture, so it can suggest the next best move for each person on a short list of high-value contacts.
19 systems
Produces and publishes short-form video and posts at scale: trend research picks topics, AI video models and a HeyGen digital twin generate the footage, Claude rewrites in the brand voice, and Blotato schedules to 14 connected accounts across seven platforms, with daily reels running unattended. Every AI video is labelled as AI-generated wherever the platform allows it.
Takes one business URL, works out what the business does and where, generates real local buyer questions, asks them across ChatGPT, Perplexity and Google's AI including the live Google AI Overview, and returns a visibility score, the competitors named instead, the exact sources AI cites and a ranked action plan. Delivered as a polished PDF report with an outreach email, and as a pay-per-use Apify Actor.
Live lead-magnet sites where a small business enters its details and gets an instant score of how often ChatGPT and Perplexity name it versus its competitors, delivered by email as a private dashboard link, with the owner notified of the lead.
Three skills run an auto-publishing AI news site: one finds news and drafts SEO-optimised articles with internal links and FAQ schema, one art-directs and generates category-specific hero images, and one audits SEO, page speed and plugin health on the generated articles only.
The research layer behind the SEO and AI-visibility work: DataForSEO for keyword volumes, trends, SERPs, backlinks, ChatGPT answer scraping and LLM-mention tracking, plus Apify for Maps and social scraping, wired into both Claude sessions and n8n.
The live multi-tenant backend behind the AI-visibility products: it onboards a business from its website, runs its buyer questions through ChatGPT, Perplexity, Gemini and Claude, records who gets named and which sources are cited, and serves scans, action plans, tiers, billing and white-label reports to members.
An agency dashboard that tracks how often ChatGPT, Perplexity, Gemini and Claude mention each client business, which sources they cite, which AI crawlers visit the site, and which directories, review sites and listicles to win next. A new client is onboarded from just a website: the site is scraped and AI generates 50 buyer prompts, competitors and industry.
Builds ready-to-post 16:9 LinkedIn carousel PDFs in a consistent hand-drawn brand style from a 90-day content plan, and scores each deck to 95/100 before delivery.
Semi-automates submitting a press release to free online PR distribution sites, using a curated site list, registration notes, a release template and a writing guide.
Scans chosen subreddits for pain points, objections, trigger phrases, emotional tone and ad angles, in 14-day chunks so trends can be compared over time; built for AI-receptionist ad copy but works for any product.
Takes any page URL, pulls its on-page SEO and the domain's ranking context, expands and enriches candidate keywords, and scores them against a difficulty ceiling scaled to the site's own authority. It outputs a prioritized content brief and a scored keyword CSV, with near-page-one wins first and then new low-competition topics.
Five chained skills take a beginner from 'I don't know what to post' to a scheduled post: capture a brand brief, brainstorm, write the post, grade it for virality with a hook-weighted scorecard and top three fixes, and schedule it through Blotato.
Applies the StoryBrand framework (BrandScript, one-liner, controlling idea, page hierarchy and campaigns) to write or audit website copy, landing pages, emails and offers across all of the network's brands.
An embedded Shopify app that reads a store's catalog, generates the buyer questions shoppers actually ask, runs them through ChatGPT, Perplexity, Gemini and Claude with web search on, and scores how often the store is recommended versus competitors, with the sources each AI cited and a trend over time.
An AI-assisted posting workflow for a local business's Google Business Profile: posts are drafted in a defined brand voice and vocabulary, approved by a person, then sent to an automation webhook that publishes immediately or queues them for a 9 AM release on their date, with a running log that prevents duplicates.
A self-serve product that scores how visible a business is inside AI answers, re-scans weekly and gives the owner a checkable action plan. It only credits a completed action with a visibility lift when the data clears an evidence gate (beyond noise, isolated from other changes, and persistent), so owners see what actually worked.
An experiment in running short-form social accounts with AI-generated personas: avatar video and image generation, scripted reels assembled with ffmpeg, a dated 90-day content calendar and scheduled posting through a social publishing API, with every decision logged in a project wiki. Every persona account carries the platform's AI-generated label.
The database foundation for a service that finds local search demand a business isn't capturing (city by trade by service), plans and generates location pages, runs SEO and conversion QA on them, publishes them, then tracks rankings, calls and leads back to each page. It ships with an 18-workflow build order and the decisions needed before production.
An admin back office that schedules LinkedIn and YouTube posts by campaign and runs them in testing or live mode with retry limits and escalation. It scores engagement with configurable weights and cuts uploaded videos into transcribed segments. n8n handles publishing and the metrics webhooks.
3 systems
Built for a Canadian consumer lender
Every minute, picks up customer payment notification emails (e-transfer and pay-by-bank), extracts the payer, amount and reference, matches pay-by-bank emails to the payment provider's API to fill in the contract number, skips anything already posted and appends new rows to a daily Excel workbook for the loan-management system, rolling over at a 5 PM business-day cutoff.
Built for a wireless and telecom engineering firm
Every hour, reads incoming purchase-order emails, takes the PO and tracking numbers from the attached PDF or the email body, finds the matching completed job on the company's Monday.com board and writes the PO number. Each email is then filed as processed or failed, replacing manual lookup and typing.
Captures the official pages and guideline PDFs of 845 Canadian business-support and grant programs, labels how usable each page's evidence is, then has an AI agent decide program by program whether a given company, or its clients, is eligible. Each decision cites quoted evidence, blockers and missing facts instead of keyword matches, and the results are validated, merged and shortlisted to the programs currently accepting technology and automation projects.
4 systems
Turns piles of reviews, comments or posts into a sentiment label and score, top themes and optional per-aspect sentiment (for example price versus service versus quality), with a roll-up across the whole set. Works on any text list or on another scraper's output.
A Postgres warehouse that pulls daily and weekly Google Analytics 4 stats for every site in the network, records first-party interaction events from a privacy-safe tracker, and keeps a hypothesis backlog and change log so marketing changes can be tied to traffic and lead movement.
Scheduled syncs pull per-post and per-video stats from YouTube, Instagram, Threads and LinkedIn into one Postgres database through a least-privilege writer role, with automatic credential renewal, so content performance can be compared across channels and accounts.
One dashboard over a website portfolio's growth program: weekly traffic with every site change marked on the chart, hypotheses with ICE scores and their evidence, on-page behaviour such as form funnels, rage clicks, scroll and exits, social channel publishing, and the decisions waiting for a person. Findings state their evidence tier instead of claiming cause from a before-and-after.
8 systems
The company's live website: service pages for AI receptionists, voice agents, missed-call recovery and business automation, town-by-town local pages and governance and capabilities pages. Every lead form posts to an automation webhook, and first-party page tracking feeds a site-intelligence database.
A static public website for an original fantasy book series: homepage, book and world overviews, character pages, concept art and a generated public wiki with client-side search. It is built from canon-managed source data (entity registry, timeline, name rules), and a build report checks for broken links and leaked spoilers or internal names.
A portfolio of about 24 domains arranged as a five-layer funnel: capture sites built around high-intent searches, agency hubs that convert, product and tool sites, a content engine, and the revenue outcomes, all served as static sites from Cloudflare's edge.
Scott Holmes' personal site: who he is, the systems he builds, a portfolio of shipped work and a public engineering work log of real bugs found and fixed. It is served from Cloudflare's edge with clean URLs and schema.org structured data for search and AI engines.
Runs a funeral home's whole case lifecycle (first call, chain of custody, arrangement, contract, statutory forms, payments) from one record entered once, and keeps working with no connectivity on phones, tablets and desktops. Provincial funeral-law requirements are enforced as declarative compliance gates, and signed contracts flow to the bookkeeper as an accounting import file instead of being re-keyed.
Built for an enterprise client proposal
The first, browser-based version of the architecture town: a Next.js and React Three Fiber world backed by PostgreSQL with versioned, event-logged scene persistence, modelled system buildings, animated data routes, governance controls, an explorer camera, a guided whiteboard tour and a building turntable lab. It also includes a deterministic image-to-asset compiler that rebuilds reference art from SHA-256-verified lossless tiles with zero pixel error.
A Chrome extension that keeps very long ChatGPT conversations fast: the browser skips work on offscreen messages and far-away turns are parked, with no history deleted. It adds whole-thread search, an outline rail, a live performance readout and a one-click 'branch this chat' that writes a compact briefing for a fresh conversation, using Chrome's on-device AI when available and making no network calls.
A multi-tenant SaaS that lets the public report non-emergency incidents (minor theft, lost property, driving complaints) online to police services and public agencies, and gives agency staff a console to review, route and export reports. Each agency is configured as data rather than code, and every staff action is written to a tamper-evident audit log.
6 systems
Fifty MCP servers covering 25 distinct services are wired into Claude so agents can act across the whole stack: automation, databases, CRM, mail, social, video, SEO data, cloud, 3D tools, a home lab, a private local LLM and a private memory layer.
Gives every project its own local PostgreSQL + pgvector knowledge base of its documents, re-indexed automatically whenever an AI coding session edits a document, and lets those sessions search one project or all of them through a single read-only MCP server. Search fuses meaning, keyword, exact-identifier and title matches, and the embedding model runs on the machine, so nothing leaves it.
Turns a repository's git history into a dated, day-by-day evidence timeline and pairs it with a short daily experiment log and an uncertainty register, so a Canadian SR&ED research claim can be written from records made at the time instead of from memory.
Audits a Reddit Devvit web app or game for production readiness, with screenshot-based visual checks at Reddit's real height-locked post sizes so clipped screens, unreachable buttons, missing assets and font failures are caught before publishing.
Audits a local static website for broken links, orphaned files, case-mismatch links that break on Linux hosts and misrouted 'Next' links, writes a report and auto-fixes the unambiguous issues; a companion skill starts and sanity-checks a project's dev server and captures the first startup error.
Applies multi-file code changes produced by an AI assistant to a Unity project from a single JSON patch file, with a dry-run preview, timestamped backups before every overwrite and encoding-safe writes.
7 systems
A free browser tool on Private5G.ca for sketching a floor plan with wall materials, placing radios and seeing a signal-strength heatmap from a free-space path-loss model with per-material wall losses, plus a 3D view. It sits alongside a set of private 5G guides by industry.
An MCP server that lets someone ask Claude for a wireless coverage plan for a named site such as a quarry, a wind farm or a point-to-point link. It finds the site, asks six questions while loading public lidar terrain in the background, and returns a conceptual design with mount positions, sector directions, coverage and handoff percentages, backhaul checks, a bill of materials and a CAD model per mount, viewable in BrowserCAD.
Built for a wireless and telecom engineering firm
Lets wireless field engineers capture a site walk with no connectivity: photos, voice notes, typed notes, GPS and elevation, against guided job checklists. It syncs when the phone is back online and drafts a client-ready survey report section by section with AI, using only what was actually captured, then exports it to Word with tracked revisions.
Turns a meeting transcript or verbal brief for a radio-frequency or wireless engineering project into a formatted scope-of-work proposal document.
Built for a wireless and telecom engineering firm
Turns phone video or photos of a building into a persistent 3D digital twin, tags walls and objects with radio materials, and simulates Wi-Fi and cellular coverage so engineers can place radios and walk through the site seeing predicted signal at every step. Reconstruction and RF are independent, so moving a radio only re-runs the RF, and opening a saved twin re-runs nothing.
Built for a wireless and telecom engineering firm
Builds a lightweight, game-style map of a building from site video, where walls, floors and ceilings carry material values that drive both the visuals and the radio model. An engineer or salesperson places access points, opens doors and changes wall materials and sees coverage recompute in about 20 ms in the browser, before any floor plan exists.
Turns a plain-language radio coverage request into constraints, places radios against them, computes terrain-aware path profiles and coverage maps with real diffraction models, grades in-building emergency radio coverage against NFPA and IFC rules, and renders a report where every RF figure carries its provenance and confidence tier. Outside AI agents can call the engine over MCP and get back only verified numbers.
8 systems
Built for an enterprise client proposal
A single 31 KB presentation page that builds an AI architecture up in 15 staged steps with presenter notes and a question to ask the room at each stage, lets the presenter isolate any part to answer questions, and lets the room draw on it with typed nodes whose connectors infer their own meaning. Drawings are checked live against the architecture's governance rules and export as a JSON diff rather than a photo.
Drops onto an existing Claude Code or AI-agent build and routes every tool call through one policy decision point that defaults to deny, confines file and shell access to a boundary, vets each skill for risk and prompt injection before it can run, and records every action in a hash-chained audit log. Published as an npm command-line tool with a resident, fail-closed policy daemon.
Puts an allow, ask-for-approval or deny gate in front of every custom function call an Azure AI Foundry agent proposes, so the call only runs once policy allows it. Approvals must come from someone other than the requester, errors and timeouts fail closed, and every decision lands in a tamper-evident audit trail. Ships as a containerized policy sidecar with Python and .NET adapters and usage metering for an Azure Marketplace container offer.
Built for a wireless and telecom engineering firm
Lets a local or desktop AI assistant read a company's NetSuite ERP within strict policy limits. Every write is only a proposal until a named person approves it with a change-ticket reference. Every read, denial, approval and change goes into a tamper-evident, signed audit log that produces auditor-ready HTML and CSV reports.
An independent, append-only ledger that records only hashes and references of events from other systems, such as AI agents or compliance apps. It chains and signs them, issues Merkle-backed certificates and offers a login-free public verification page, proving what happened and that the record was not altered without holding the business data itself.
Moderates new forum threads and posts in layers: browser-side checks, a database pattern pass, then a cost-capped Claude review triggered from the database. The AI reviewer understands that a security-education forum may discuss prompt injection, and only flags real attacks, abuse or spam, returning a structured verdict.
A desktop app for launching, watching and approving the work of governed AI agents. It embeds the same governance core and live policy daemon, so every agent action on screen has already passed the deny-by-default gate, and irreversible operations need the operator's confirmation.
A machine-checked build plan for an engine that stands up a client-owned AI and automation platform, with dev, test, UAT and production all in the client's own Azure subscription, certification gates, client-approved promotion and a client-owned support lane. Today it is the plan pack plus tooling that proves every proposal requirement traces to a deliverable, a work package and a test.
6 systems
A CAD workspace that runs in the browser for telecom site design: open DXF or DWG site drawings, build towers, mounts, antennas, radios, cabinets and cable runs from a parametric library, run live design checks, and export antenna schedules, CSA S37 wind-area tables, DXF and 11x17 PDF sheets. A network view checks wireless links against national lidar terrain and explains blocked paths in plain language.
Built for an enterprise client proposal
A small explorable 3D town that explains a proposed agentic AI platform to a client: systems are buildings, data moves as postal vehicles, human approvals are toll gates that hold traffic until a person decides, and process stages glow as an overlay. A built-in spec checker refuses to render any layout that breaks the governance rules, such as making governance a building you could route around. Runs on the desktop and as a web build.
An interactive 3D anatomy module for students: pick an event, such as a pinprick on the foot or a rise in blood pressure, and watch the signal travel through a rotatable, zoomable, explodable 3D body from receptor to cortex, including the reflexes and outputs that branch off along the way.
Claude is wired directly into Blender, Unity and Godot and into Meshy's text- and image-to-3D service, so scenes, assets and game builds can be generated, inspected and iterated from a conversation across the 3D, RF-twin and game prototypes.
Built for a wireless and telecom engineering firm
Converts scanned, raster-only architectural floor plans into layered CAD files for RF coverage modelling: the building envelope, interior walls and elevator or stair core are extracted by image processing, workstations can be placed as real CAD blocks, and each floor is released as versioned DXF and DWG with a per-floor quality status. The envelope step is packaged as a reusable floor-plate outline skill.
Built for a wireless and telecom engineering firm
Turns an RF readings spreadsheet, site photos and a PDF floor plan into an organized AutoCAD-ready drawing: the plan is traced at true scale, every reading is plotted and colour-coded by signal strength, photo markers are placed and filed, and layers, a legend and a QA summary are stamped on, so an engineer only verifies and polishes instead of drafting from scratch.
3 systems
Distils hundreds of past AI work sessions into a project-organized, evidence-grounded wiki that a scheduled task re-reads and corrects daily, with a session index, tags, backlinks and entity pages. AI assistants read it through a zero-dependency MCP server that is read-only by construction: it cannot write, delete or run anything, and blocks any path outside its folder.
Built for an electric utility's wireless network rollout
Turned 18 months of project email metadata (about 9,700 messages), 650 meetings, transmittals, trackers and design-review documents from a multi-site utility wireless network replacement into a redacted, annotated, searchable knowledge base. It then mines that base for review-cycle times, SLA performance, post-approval revisions and root-cause patterns, presented as executive dashboards.
Built for a wireless and telecom engineering firm
Runs an open-weight LLM entirely on a company's own server, without a GPU, and answers questions from its project documents with numbered citations. Documents are mirrored read-only from the company's cloud file store and embedded into a local vector database. The database itself keeps financial content out of customer-facing answers, so it doesn't depend on prompting.
1 system
Module one of a construction intelligence platform: it ingests building permit applications (forms, PDFs and IFC building models), extracts and normalizes the facts, and runs deterministic, cited building-code rules that return Meets, Does not meet, Information not available or Uncertain. Human reviewers keep final authority, AI may assist but never approves or rejects, and every finding is traceable to its source in an append-only, checkpointed audit ledger.
5 systems
Reads a student's course outlines from PDFs, spreadsheets and photos, extracts every deadline, weight, grading rule and undated obligation, plans the work backwards across the term and opens each morning to one ordered list of what to do today, rebuilding the plan when a day slips. An installable web app live at fazoozle.com as an invite-only beta.
A live learning site at prysio.com that turns thousands of real n8n automation workflows into step-by-step lessons, organised by skills, career paths, personas and business niches. Learners get animated workflow playback, a guided build mode, quests and XP, sign-in and a community forum, and can ask for any workflow to be built for them.
Turns a one-sentence request into a finished tutorial video: it builds the real workflow in n8n, captures editor screenshots, writes the narration, renders a HeyGen digital-twin intro and outro with cloned-voice walkthrough audio, and assembles the MP4 with ffmpeg.
An installable, fully offline phone app that teaches trade math to a machinist apprentice: 13 topics and about 8,300 generated questions, one-question drills with feedback that names the specific mistake, and 11 shop calculators that show their working from the same tested math engine. Built to an ADHD-first brief: short sets, no timers, a no-shame option to park a question, and streaks that survive skipped days.
An offline, single-page practice engine with 992 original scenario questions for Microsoft's 2026 Azure AI developer certification track (AI-103, AI-200 and AI-300), including case studies, ordering questions and a filter that drills only exam-style items.
18 systems
A live arena at mcpagentwars.com where autonomous AI agents join over the Model Context Protocol and fight, loot and survive against each other and house bots. An agent's gear is its tool list, so picking up or losing an item adds or removes the actions it can call; humans can only watch.
A daily Reddit game where the right answer is whatever the crowd picks: players lock in the option they think most people will choose plus a rare one they bet almost nobody picks, then a game-show style reveal shows how closely they think like the hive. A scheduled job posts a new prompt every day, with streaks and leaderboards.
A gamified road-trip app for passengers: log roadkill and roadside oddities with one tap, earn stickers, points, streaks and badges, and see every find on a shared public map. It works fully offline on long drives, syncs when signal returns and installs to a phone's home screen.
A Dwarf Fortress-inspired colony management game that runs in the browser with no install. Players dig, mine, haul, build workshops and keep a fortress alive across depth levels. A shared public world runs the same deterministic simulation on the server, so several players can join one live fortress.
A persistent medieval world that runs inside a Reddit post. A 72-region continent advances once per real day for every player, and each player spends one meaningful action a day as an adventurer or a noble, with permadeath and legacy. The server is fully authoritative and strips every hidden number from its responses, so players see words, never stats.
A 3D tower-defense game where enemy fleets attack a capital ship from every direction, built around an orbit camera, a radar scope that tracks off-screen threats, a research tech tree, custom turret loadouts and crafting from component drops.
A coding game that runs inside a Reddit post: players write small programs in a custom scripting language to drive ants that forage, dig and farm, watch the server's deterministic simulation play back, and spend what they earn on upgrades that unlock new language features, with per-post leaderboards and daily community goals.
A server-authoritative four-player fighting game inside a Reddit post, with turn-based and real-time modes, roguelite boon drafting, and bots that fill empty or idle seats so a match never stalls, plus a per-community leaderboard.
A wave-survival roguelite where you steer a segmented caterpillar whose body parts carry weapons: coil to dodge, uncoil to lash, lose segments and recover them, and molt to rebuild. Built test-first with headless assertions and an automated bot play-through, and exported to Windows and Android.
A small idle fishing game that sits in a transparent, borderless, always-on-top window in a corner of the desktop and fishes by itself, including catch-up while closed. Players sell catches to unlock five fishing spots and buy gear, and the window keeps its size and position correctly across monitors with different scaling.
An idle auto-battler where eight dwarf heroes fight waves of monsters while a Factorio-style workshop of mine, smelter, forge and conveyors makes their gear. One shared game logic ships as three single-file builds with different renderers (pixel-art canvas, a Three.js 3D diorama and a vector chibi style) plus the original Godot build, all sharing one save.
A cozy idle and care game that runs inside Reddit posts: adopt a space rock, care for it, earn currency and evolve it through eight tiers via a 30-line upgrade tree, 50 cosmetics, daily quests, leaderboards and a daily community goal, with all game state held server-side.
A cyberpunk idle/incremental game: the player builds a fleet of income-generating daemons, runs long raids into six corporate networks, fights turn-based boss battles and resets through two prestige layers. It was built largely in autonomous AI build windows, with a headless test suite and a balance-simulation bot checking every change.
A single-player puzzle played entirely through MCP: the agent starts with look, listen and touch, none of which open the door, and the only way to win is to make a tool exist that was not there at the start. It explores deceptive tool descriptions and schemas as specifications.
A Reddit colony-survival game where a dwarven crew digs into a mountain while balancing air, food and morale that feed into each other; everyone in a subreddit digs the same persistent mountain each week, with a homestead above ground and a weekly leaderboard. The game is a pure, deterministic simulation balanced by a bot that plays the real rules.
A data-driven fantasy RPG prototype in Godot where parties run missions, travel routes, craft and fight while the player is away, with deterministic combat, an offline progress simulator and versioned saves. Game content (887 items, 245 class skills, 52 registries) is validated against JSON schemas, and AI agents search the design docs through the project's own graph-memory MCP server.
A side-scrolling action RPG in development that pairs deep ARPG systems (loot, affixes, skills, passives, bosses) with readable 2.5D combat. Characters are AI-generated 3D models that Blender renders into layered sprite sheets, so each piece of gear a character equips shows on them at runtime.
An early Unity prototype that combines grid-based tower defense with a resource economy: building placement on a grid, a resource inventory, enemies spawning and following a path, and towers targeting them on a shared game tick.
330 technologies across the systems above. The number shows how many systems use each one. Machine-readable version: systems.json.
Catalogued on 2026-10-06 from source code, deployments and platform records. Counts include only systems we could trace to real code or a live deployment.
Tell me what you need. I will show you a working prototype in 5 business days, free, with no obligation.
Get a Free Working Prototype