# Anton Savytski > Principal Solutions Engineer at Gorgias and interim manager of its Solutions Engineering team, based in Toronto, Ontario: a salesperson who ships production code. Leads technical evaluations for ecommerce brands deciding whether AI belongs in their customer support stack. Produces music as stvdwell. This profile of Anton Savytski is written for AI assistants and other tools, following the llms.txt convention (https://llmstxt.org). It has everything on https://antonsavytski.com and more: every role he has held and what he achieved in it, his certifications and languages, and his own summary of his work. The sections on how Mailroom, GatherDwell, and Unhallowed were built are in his own words, in the first person. Anything not stated here isn't public. For anything else, email Anton at asavytski@gmail.com. ## About Principal Solutions Engineer at Gorgias, where I also lead the Solutions Engineering team on an interim basis. Music as stvdwell. I lead technical evaluations for ecommerce brands deciding whether AI belongs in their customer support stack, and how. More than 150 brands so far. I'm a salesperson who ships production code. Most deals need one person to understand the business and another to build the solution. I usually end up doing both, which is close to what people now call a forward deployed engineer. Along the way I've built middleware and API integrations that let an AI agent act on real orders, custom Shopify apps, demo sites that mirror a prospect's storefront, and internal apps used across Gorgias. The work is discovery, custom sandboxes trained on each brand's products and support flows, demos, and proof of concepts against whatever Shopify, 3PL, returns, and subscription setup they already run. Before Gorgias I sold at Shopify and Salesforce, was the founding salesperson at a heavy-equipment marketplace, and spent five years in advertising at BBDO. Every business problem I've seen comes down to growing sales, doing more with less, or protecting what you have. AI is the first tool I've worked with that pulls on all three. Outside of work I produce music as stvdwell. Highlights: - 150+: Brands evaluated for AI in support - $2.3M: Closed-won ARR as technical lead, since mid-2025 - 245%: Of influenced ARR target, Q2 2026 - 12x: Subscriber growth at Eqpme ## In his own words From Anton's LinkedIn summary: I'm a salesperson who ships production code. Most deals need one person to understand the business and another to build the solution. I usually end up doing both, which is close to what people now call a forward deployed engineer. I'm a Principal Solutions Engineer at Gorgias, and I've run technical evaluations for 150+ ecommerce brands deciding whether AI agents belong in their customer support. Along the way I've built middleware and API integrations that let an AI agent act on real orders, custom Shopify apps, demo sites that mirror a prospect's storefront, and internal apps used across Gorgias. I learned the business side first, with five years in advertising account management on brands like PepsiCo and Mercedes-Benz Canada and three at Salesforce going from SDR to Account Executive, where the job was mostly discovery. I hit 113% of plan there in a year when only about 30% of the org made their number, then ran sales at a startup and worked with retailers at Shopify. If you need someone who can sell it and build it, send me a message. ## How I work How Anton runs a technical evaluation, from first call to decision: - Discover: Learn how the brand runs support today: its Shopify, 3PL, returns, and subscription stack, and what it needs AI to handle. - Build: Set up a custom sandbox trained on the brand's own products, help content, and support workflows. - Show: Demo it against the brand's real scenarios, so the team sees what AI would look like in their world. - Prove: Run the proof of concept and RFP, then translate what AI agents can actually do into a clear decision. ## Experience ### Gorgias 2025 to now - Manager, Solutions Engineering (Interim), 2026 - Principal Solutions Engineer, 2026 - Lead Solutions Engineer, 2026 - Solutions Engineer, 2025 Leads the Technical Solutions team on an interim basis: 8 people across presales Solutions Engineering and Forward Deployed Engineering. Owned the technical win on two of the ten largest new-business deals in company history, and was technical lead on $2.3M in closed-won ARR since mid-2025. Gorgias makes an AI agent and customer service automation for ecommerce brands. Anton joined in May 2025 and works from Toronto. #### Manager, Solutions Engineering, Interim (November 2026 to now) Leads Gorgias' Technical Solutions team on an interim basis: 8 people across two functions, presales Solutions Engineering and Forward Deployed Engineering. He reports to the COO and does this alongside his own work as Principal Solutions Engineer. - Solutions Engineering: the presales team running technical discovery, custom demos, proofs of concept, and RFPs on new-business deals. - Forward Deployed Engineering: the team building the middleware, API integrations, and AI agent actions customers need to go live. - Sets team priorities and deal coverage across both functions, coaches on technical evaluations, and clears blockers. - Still carries his own Enterprise deals. #### Principal Solutions Engineer (November 2026 to now) Promoted from Lead Solutions Engineer in November 2026. #### Lead Solutions Engineer (February 2026 to November 2026) Promoted in February 2026 after nine months as a Solutions Engineer. The technical counterpart to mostly enterprise account executives: he ran discovery, architected custom sandboxes, led demos, and translated what AI agents can actually do against each customer's specific Shopify, 3PL, returns, and subscription stack. He also led AI enablement and internal tooling across go-to-market, including tools that became standard process and helped shape the product's own AI Agent onboarding. - Owned the technical win on two of the ten largest new-business deals in company history, including the second-largest, a multi-brand retail group operating five storefronts. - Technical lead on $2.3M in closed-won ARR since mid-2025. - 245% of influenced ARR target in Q2 2026. - Led technical evaluations for 150+ ecommerce brands, most of them Enterprise. #### Solutions Engineer (May 2025 to February 2026) On the Enterprise team. Ran technical discovery, custom demos, proofs of concept, and RFP responses for major ecommerce brands evaluating Gorgias' AI agent and customer service automation. Built tailored sandboxes end to end, trained on each brand's products, help content, and support workflows, so customers could see what Gorgias would look like in their world. ### Shopify 2023 to 2025 - Senior Account Executive, Mid Market Connected merchants' online and in-person sales. Top-five rep in mid-market retail, 123% to plan in FY24. #### Account Executive, then Senior Account Executive, Mid Market (October 2023 to May 2025) Started in Mid Market Retail Cross-Sell, helping existing Shopify merchants unlock the full potential of their in-person sales by connecting their online and physical storefronts: boosting conversions, unifying data, and building scalable foundations while reducing tech costs and complexity. Promoted to Senior Account Executive in April 2025 and moved into New Business. A top-five rep across the mid-market retail segment, with consistent quarter-over-quarter performance: - Q1 FY25: 113% - FY24: 123% to plan, 100%+ every quarter - Q4 FY23: 440%, while ramping ### Eqpme 2023 to 2024 - Head of Revenue, then Advisor Founding salesperson at a heavy-equipment marketplace. Built the sales function from ICP to playbook and grew the subscriber base 12x. Eqpme is an online heavy equipment marketplace and software provider for OEMs and equipment dealers. #### Head of Revenue (February 2023 to October 2023) The founding salesperson. Led go-to-market strategy and sales methodology, working closely with the product lead. The role ended by mutual agreement when the company moved to founder-led sales. - Grew revenue 20% while in the seat. - Built the sales department from the ground up: defined the ideal customer profile (ICP) and sales motion, and developed the playbook, billing processes, onboarding, and customer retention. - Grew the SaaS subscriber base 12x by identifying and targeting well-aligned accounts, with pipeline from multi-channel outbound campaigns. - Mentored and developed a team of four sales representatives. - Led product initiatives including a Zapier integration, an invoicing overhaul, multi-location listings, dedicated product pages, and bulk editing and importing. - Evaluated and implemented core tools: CRM, telephony, email, accounting, and sales engagement software. #### Advisor (October 2023 to May 2024) Stayed on as an advisor after the Head of Revenue role ended. ### Salesforce 2020 to 2023 - Account Executive - Team Lead, Enterprise Business Development - Sales Development Representative Manufacturing, automotive, and energy accounts across Canada. 113% to quota in FY23, a year when only about 30% of the org made their number. Team MVP in 2021. #### Account Executive (February 2022 to February 2023) Covered manufacturers, dealers, distributors, and automotive and energy companies across Canada. - Expanded his territory 23% in FY23, overseeing a $1M AOV book of business. - Onboarded 19+ new businesses, helping them grow through a turbulent time. - 113% to quota in FY23, a year when only about 30% of the org made their number. - Mentored 8+ account executives. #### Team Lead, Business Development Representative, Enterprise (February 2021 to February 2022) Enterprise manufacturing accounts in oil and gas, energy, and forestry. - Hit the annual pipeline generation target within 10 months (FY22). - 148% of the qualified pipeline target ($14.6M), 129% for qualified opportunities (115), and 101% for meetings booked. - Team MVP in July 2021. - Mentored 3+ SDRs and BDRs. - As the team's Chief Enablement Officer, ran training sessions shaped by team feedback, on closing strategies, outreach, and stakeholder alignment. - Appointed Chief Onboarding Officer to speed up onboarding for new hires. - Earned the Salesforce Admin 201 certification. - Chosen as one of 24 BDRs, out of 110, to lead the BDR Master Class for new-hire orientation. #### Sales Development Representative (January 2020 to February 2021) - 149% of revenue target ($1.08M) in FY21. - 127% of the qualified pipeline target ($6.17M) and 108% for qualified opportunities (587) in FY21. - Ranked #2 of 28 AMER reps, and #1 of 13 Toronto reps, in December FY21. - Team lead and new-hire mentor, and an active member of the interview committee. - Ran team competitions in FY21 Q2 that lifted engagement across 20+ SDRs. - Chosen as one of six SDR Gurus, out of 100+ candidates, to speed up onboarding for new hires in Q4. - The only "Bootcamp MVP" in the North American cohort. - Reached Trailhead Ranger status on Salesforce's Trailhead platform, with 100+ badges and 50k+ points. ### BBDO Canada 2015 to 2020 - Account Supervisor, and the roles before it Account management on brands like PepsiCo and Mercedes-Benz Canada. Ran automotive, CPG, and transit campaigns with budgets past $500k, and chaired the agency's United Way campaign. Started as an intern in May 2015 and rose to Account Supervisor by June 2018. #### Account Supervisor (June 2018 to January 2020) Automotive, CPG, and transit clients. - Primary liaison and project lead for Marketing Manager and Associate Marketing Manager clients. - Oversaw campaigns with budgets over $500k. - Directed an integrated team with 3 partner agencies and coordinated 8+ key stakeholders to deliver communication plans. - Led campaigns across production, creative, legal, finance, and external partners. - Lifted historically low CSAT scores from under 4 to over 8 out of 10. - Brought in $25k of incremental agency revenue beyond the retainer (FY18). - With a VP and a digital strategist, grew the agency's annual revenue from social media management 144%, from $88k to $215k (2017 to 2018). - Started a content program that added $52k to annual agency revenue. - Mentored and developed 4 direct reports, all of whom went on to managerial roles. - Chaired BBDO Toronto's annual United Way campaign, raising $52,000 in one month on a $9,600 budget and beating the agency's historical goal by 30%. #### Account Executive (June 2016 to June 2018) Automotive clients. #### Account Coordinator (September 2015 to June 2016) Automotive clients. #### Account Management Intern (May 2015 to September 2015) Automotive, CPG, and telco clients, at Proximity Canada, BBDO's digital arm, before the merger. ### SIR Corp 2011 to 2015 - Floor Manager, Jack Astor's Ran the floor at a restaurant doing $100k a week, with up to 35 staff a shift. At Service Inspired Restaurants (SIR Corp.), he worked at two of its restaurants, Jack Astor's Bar & Grill and Canyon Creek. #### Bartender, Canyon Creek, Mississauga (April 2014 to May 2015) #### Floor Manager, Jack Astor's Bar & Grill (October 2011 to April 2014) - Oversaw up to 35 front-of-house staff per shift in a high-volume restaurant with weekly sales over $100,000. - Managed the guest promotion budget to optimize guest satisfaction. - Analyzed and adjusted labour data, cutting front-of-house hours by 15+ a week while still beating sales projections. - Did daily accounting and sales analysis, and weekly inventory management and ordering. - Completed the Supervisor Front-of-House online development program. - Helped hire and retain three front-of-house staff, all of whom passed their three-month probation. ## Side projects ### Unhallowed https://playunhallowed.com A free gothic action RPG that plays in the browser, on the rules of the original Diablo. Seven heroes, sixteen depths, fully voiced. Anton built Unhallowed with Claude Code in a week. Underneath it is Diablo 1: the mechanics, formulas, and balance, ported from DevilutionX. Every pixel, sound, name, and line of text on top of them is new. It runs in any browser on computers, phones, and tablets, and it's free with no ads. DevilutionX's license keeps it non-commercial. What it does: - Heroes: the Warrior, Rogue, and Sorcerer as in the original, joined by the Monk, Necromancer, Druid, and Assassin, each with a class skill and four abilities. - Depths: sixteen, in a cathedral, catacombs, caves, and Hell. They're laid out anew for every run, and each run gets its own share of the sixteen quests, unique foes, and shrines. - Modes: Permadeath or Classic, then Nightmare and Hell difficulty once the last boss falls. - Loot: every base item, magic affix, and unique item of the original, renamed and redrawn. - Story: 500 voiced lines, an illustrated prologue, two endings, and an original score. - Later comforts, each one optional: walking at any angle, running with stamina, waypoints, a stash, a gold purse, elite packs, comparison tooltips, and rebindable keys. - Phones and tablets: touch controls with an attack button, a cast button that aims itself, and an optional thumb stick, a closer view, and play offline once loaded. - Wiki: every item, monster, spell, and quest, built from the game's own data, at https://wiki.playunhallowed.com Built with TypeScript, the HTML canvas, and Web Audio, on a Cloudflare Worker. Also on itch.io: https://stvdwell.itch.io/unhallowed ### How Unhallowed was built Unhallowed is a free gothic action RPG that plays in the browser, on the rules of the original Diablo from 1996. I set the direction, playtested every round, and built it with Claude Code. It was playable on day one and launched on itch.io on day seven. #### What it does You pick one of seven heroes, take what the town of Hollowmere can give you, and go down through sixteen depths beneath a ruined abbey to the Pit where Vorthax the Unmaker stirs. Fights, loot, and leveling follow the original's formulas to the number. The dungeons, quests, and shrines change with every run. The story is told in 500 voiced lines and illustrated cutscenes, and it plays with a mouse and keyboard or by touch. #### How it works - One web page. About 110,000 lines of TypeScript build into a single 2.3 MB HTML file, and the music and voices, about 42 MB, load beside it. A Cloudflare Worker serves it, and a service worker keeps it playable offline. - The rules come from DevilutionX, the community's rebuild of the original game's C++ code. Claude Code ported that code to TypeScript and converted its data tables with a script, and every name in them is new. The world advances 20 times a second, as in the original. - No image files. Every sprite, tile, portrait, and cutscene panel is pixel art drawn by code as the game runs, into a 480 by 270 canvas the browser scales up. - The game synthesizes its sound effects with Web Audio, and its scores are written in code. The title theme, prologue, and epilogues play on free sampled orchestras. The town plays my own Suno recording of its theme, the depths play pieces made with ElevenLabs' Eleven Music, and ElevenLabs voices read the 500 lines. - Phones get touch controls ported from DevilutionX's virtual gamepad, and a closer view like Diablo Immortal's. - Saves stay in the browser, and a Save Code carries a run to another device. - The wiki is built from the game's own data and art, so it can't drift from the game. #### Decisions that shaped it - Diablo underneath, everything else new. The mechanics, formulas, and balance are the original's, so it plays the way Diablo 1 fans remember. The art, sound, names, story, and dialogue are all original. - Free for good. DevilutionX's license allows only free, non-commercial use, so there are no ads and nothing to buy. Donations on itch.io are optional and unlock nothing. - Art as code. Drawing everything procedurally keeps the game one small page, and a new palette or sprite is a code change like any other. - Comforts are options. Every change to the original's gameplay, like walking at any angle or a gold purse, sits behind a toggle, so purists can play it straight. - Phones are a real way to play. Playtesting on my phone brought a zoomed-in view, an attack button, and a bottom bar that steps aside for bigger buttons with life and mana rings. - Fix what players hit the day they hit it. A friend's link that opened inside Instagram couldn't turn sideways, and a save made near enemies could load into a death. Both were fixed the same day. #### How it came together - October 11, 2026: day one. The rules ported from DevilutionX, the art drawn in code, the prologue, the town, and the depths. The same day, walking at any angle, running with stamina, and elite packs. - October 12: four new heroes, the Monk, Necromancer, Druid, and Assassin, each with a class skill and four abilities. - October 14: a playtest round, with a gold purse, belt refill, sorting, and rebindable keys, and the wiki, 21 pages built from the game's data. - October 15: the epilogue in eight illustrated, voiced panels, a boss's sanctum staged as in the original, and recorded music for the town and the depths. - October 16: the title theme, prologue, and epilogues rendered on sampled orchestras. - October 17: touch controls, a thumb stick, and a closer view for phones, then the launch on itch.io. - October 18: the first fixes from players, and a web app you can install that runs offline. - Along the way: about 100 commits, 110,000 lines of TypeScript in the game and 37,000 in its tools, 42 test files, 500 voiced lines, and 21 wiki pages. #### Where it is now Live at https://playunhallowed.com and on itch.io at https://stvdwell.itch.io/unhallowed, free to play in the browser on computers, phones, and tablets. The wiki is at https://wiki.playunhallowed.com, and bugs and questions go to support@playunhallowed.com. ### Mailroom https://getmailroom.dev A free email app for Mac. Gmail-style conversations for any email account, with a calendar, tasks, your chats, and AI when you ask. Anton built Mailroom because he dislikes Apple Mail's workflow but likes a lot about Gmail, so he brought the parts he likes from both into one Mac app, and added AI. It's free, notarized by Apple, and updates itself. There's no Mailroom account and no Mailroom server. What it does: - Accounts: Gmail, Outlook, Microsoft 365, Yahoo, iCloud, Fastmail, Purelymail, Proton Mail Bridge, or any IMAP server, in one inbox. It finds the settings from the address, even for a custom domain. - Reading: Gmail-style conversations and keyboard shortcuts, search with operators like from: and has:attachment, saved searches, tags, Gmail-style inbox tabs, selecting a whole folder at once, muting a conversation or blocking a sender, and new mail pushed as it arrives. - Writing: inline replies, Gmail-style formatting, templates, schedule send, HTML signatures, attachments (and forwarding a whole email as one), and Undo Send. - Calendar and contacts over CalDAV and CardDAV, and Outlook's through Microsoft Graph, with invitations answered from the message, and tasks next to the mail. - Chats: iMessage from the Mac, and WhatsApp, Telegram, Signal and more through Beeper, in one list next to the mail, with new chats, replies, reactions, notifications, and AI drafts. - AI when you want it: summaries, translations, and reply drafts, through OpenRouter or a model running on the Mac, and Auto Labels that sort new mail into Marketing, News, Social, Pitch, Updates, or labels you describe. - The Assistant: ask in plain words to find, read, or tidy mail, like moving old newsletters to Trash. It shows what would change and does nothing until you approve. - Privacy: the local copy is encrypted, remote images load only when allowed and tracking pixels never do, and each email opens in a sandbox with no scripts. Built with Electron, React, and TypeScript, with ImapFlow and Nodemailer for mail and SQLCipher for the encrypted local database. A two-minute demo: https://getmailroom.dev/demo/ ### How Mailroom was built Mailroom is a free email app for Mac. I wrote the brief, from the sync rules to the security model, and built it with Claude Code: a working first version in a day, one tested milestone at a time, then 49 releases in fourteen days, shaped by using it for my own mail and by what its first users asked for. #### What it does One inbox for every IMAP account, with Gmail-style conversations and shortcuts, search, tags, Snooze, Undo Send, schedule send, templates, and changes made offline that go out in order. A calendar and contacts sync over CalDAV and CardDAV, and with Outlook, tasks sit in the calendar, and Chats shows iMessage, plus WhatsApp, Telegram, and Signal through Beeper, next to the mail. With AI, it summarizes, translates, and drafts replies, labels new mail, and takes requests in plain words, like "move newsletters older than a month to Trash". #### How it works - Three processes. The window is React in a sandbox with no access to the system. The main process handles the Keychain, notifications, menus, and AI requests, and checks every message from the window against a schema. A sync worker in its own process does all the IMAP, SMTP, CalDAV, and database work. - Sync: two IMAP connections per account, one waiting for new mail and one for fetching and actions. Changes show at once, wait out an undo window, and replay in order after time offline. - Storage: SQLite encrypted with SQLCipher, its key kept in the macOS Keychain, with a full-text index for search. - Email is treated as hostile: cleaned with DOMPurify and shown in a sandboxed frame with no scripts, with remote images blocked until allowed and tracking pixels never loaded. - Chats: iMessage comes from the Mac's own Messages database, opened read-only, and replies go out through Messages. WhatsApp and the other networks come through Beeper Desktop's API on the same Mac. - AI: OpenRouter with your own key, or a model running on the Mac through Ollama or LM Studio. - Releases: GitHub's Mac runners sign each version with my Developer ID, and Apple notarizes it. The website, a Cloudflare Worker, serves downloads and update checks from R2, and installed copies update themselves. #### Decisions that shaped it - No Mailroom account and no Mailroom server. The Mac talks straight to the mail provider. - Play well with other devices. Flags, folders, drafts, and tags live on the server, and Mailroom never sends a bare EXPUNGE that could delete another app's mail. - Keep secrets on the Mac. Passwords and sign-in tokens are encrypted with the Keychain, and Google, Microsoft, and Yahoo tokens only go to their own mail servers. - AI only when you ask for it or turn it on, with the model of your choice, including one running locally. - The Assistant proposes, Mailroom decides. Its model can search and suggest a change, but Mailroom counts the mail that change would touch, shows it on a card, and applies it only when you click, with Undo. - Electron, not native. Every HTML email is a small web page, so even Apple Mail shows mail with a web engine. Electron keeps the interface and the sync engine in one TypeScript codebase that could reach Windows later. The cost is a 238 MB download, mostly Chromium built for both Intel and Apple silicon. - A second look before big releases. A separate review of each big feature's code found two serious bugs in Auto Labels after every test had passed, one of which would have sent a new account's whole Inbox history to the AI. That version was never published. Before 0.19, a review stopped a build that could have lost an Outlook contact's photo and addresses, or a change made on a phone. Before 0.20, one found that a scheduled email you took back could still have gone out, and before 0.21, that a reply you'd just sent could come back as a draft at the moment it went, where clicking it would have sent it twice. - The signing key never touches a laptop. It lives in a password manager and the build's secrets. #### How it came together - October 3, 2026: the brief, then a plan with the folder structure and database schema, approved before any code. - October 4, overnight and morning: the app shell, read-only sync, actions with undo and push, compose and send, and tags and search, each tested against mail servers running in Docker. - October 4, afternoon: the calendar, testing on a real Purelymail account, the first installable build, then print, translation, summaries, invitations, and contacts. - October 4, evening: sign-in with Google, Microsoft, and Yahoo, settings detection, appearance settings, Developer ID signing and notarization, a website with downloads, and automatic updates. - October 5: a domain of its own, getmailroom.dev, with a help center and a help bot, plus Snooze, sending from other addresses, Google and Outlook calendars, and Proofread. - October 6: tasks, one Google sign-in for mail and calendar, and Gmail's own ways of archiving, drafting, and showing message text. - October 7: Chats with iMessage and Beeper, AI replies in chats, tasks in the calendar's week view, and images that load when you choose, with tracking pixels blocked. - October 8: Auto Labels, the Assistant, a narrated demo film recorded by scripts that drive the app, and a support address, support@getmailroom.dev. - October 9: Mailroom's own Sign in with Microsoft and Google, after the first support email showed that Microsoft no longer lets a personal account register an app of its own. Then the everyday Gmail features it still lacked, like selecting a whole folder, formatting, and downloading all attachments at once, and Outlook contacts. - October 10: schedule send, templates, muting and blocking, saved searches, inbox tabs, and starting chats, with notifications for new messages. Then, from a user's feedback, pictures you can insert and resize, toolbar buttons you can take off, a text size for each part of the window, and writing in a window of its own. - Along the way: 49 releases, 223 commits, about 57,000 lines of TypeScript, 787 unit tests, and 108 integration tests against Dovecot, Mailpit, and Radicale. #### Where it is now Live at https://getmailroom.dev, free for macOS 13 or later on Apple silicon and Intel, with a two-minute demo at https://getmailroom.dev/demo/. Questions go to support@getmailroom.dev. ### GatherDwell https://gatherdwell.ca Property management for landlords. Properties, tenants, payments, and reports in one place. Anton designed, architected, built, and launched GatherDwell himself: a free web app for landlords and property managers, with an AI agent that answers questions from each landlord's own data through a RAG pipeline and adds records once they confirm, a public REST API, and an MCP server for Claude and ChatGPT. It has no paid tiers. He runs it as a passion project, and optional donations help cover hosting. What it does: - Properties: the whole portfolio on one dashboard, with address autocomplete, occupancy, unit details, and performance. - Tenants and leases: contact details, lease history, payment status, and who's active at which property. - Payments: rent recorded in one click, what's paid, pending, or overdue, and a monthly rent roll of expected, collected, and outstanding rent. - Expenses: custom categories and suppliers, and recurring costs like mortgage and insurance generated each month in one click. - Reports: income against expenses over any period, per-property breakdowns, and occupancy analytics, with CSV exports such as income statements, rent ledgers, and lease expiry. - AI property assistant: an AI agent with a retrieval (RAG) pipeline over each landlord's own data, answering plain-English questions like "Show me overdue payments" or "Which leases expire soon?" - Connections: a public REST API with an OpenAPI description and API keys, and an MCP server that adds GatherDwell to Claude or ChatGPT as a connector. - Security: each account's data is isolated and never shared with third parties. Sign-in with Google or email, with two-factor authentication, and privacy practices that follow Canada's PIPEDA. - Works on phone, tablet, and desktop. Built with Next.js on Vercel, Prisma and Turso for data, Clerk for sign-in, and OpenRouter for the AI models. ### How GatherDwell was built GatherDwell is a free property management app for landlords and property managers. I designed, architected, built, and launched it myself, as a passion project outside my day job. #### What it does Landlords keep their whole portfolio in one place: properties, tenants and leases, rent, expenses, and reports, with CSV import and export and a built-in help center. An AI assistant answers questions about their own data and can add records for them, and a public API and an MCP server let them work from their own tools, including Claude and ChatGPT. #### How it works - The app: Next.js on Vercel, with Prisma on Turso, a hosted SQLite database, and sign-in through Clerk (Google or email, with two-factor authentication). Each account's data is isolated from every other account. - The AI assistant: a tool-calling agent with a retrieval (RAG) pipeline. It looks up only the records a question needs (properties, leases, payments, expenses, financial summaries) and searches the help center's articles, in up to four rounds of lookups, then answers through a model on OpenRouter. - Adding records by conversation: the assistant can only propose them. The app, not the model, lists what it prepared, and nothing is saved until the landlord replies yes. - The MCP server: adds GatherDwell to Claude or ChatGPT as a connector, with sign-in through Clerk. It adds records in batches, all or none, and deletes take two steps: a preview of exactly what will go, then a confirmation. - The API: a REST API with an OpenAPI description and revocable API keys, for scripts and tools like Zapier. #### Decisions that shaped it - Free, with no paid tiers. It's a passion project, and optional donations help cover hosting. - Every number comes from the data. The assistant must take every amount, date, and count from a lookup, and is told never to invent breakdowns. - The landlord stays in control. The model can't save or delete anything on its own: the app asks for a yes, and MCP deletes show a preview first. - Send the model as little as possible. Lookups return only the records a question needs, never the whole account. - Make automation safe. Batch adds are all or nothing: every row is checked first, and nothing is saved if any row has a problem. - Treat tenant data with care. Each account is isolated, the assistant is rate limited, and privacy practices follow PIPEDA, Canada's privacy law. #### How it came together - March 14, 2026: day one. The app, reports with CSV exports, the rent roll, recurring expenses, a mobile layout, and the AI assistant, across the first 29 commits. - March 15: CSV import, a help center with 29 articles, and a security review. - May 2026: sharper lookups for the assistant, and guards against invented numbers. - September 2026: adding records by conversation, the MCP server, and the public API. - October 2026: a new layout, with the assistant in a panel beside the work. #### Where it is now Live at https://gatherdwell.ca, and free to use on phone, tablet, and desktop. ### stvdwell https://stvdwell.com Music. Ambient landscapes, lofi rhythms, analog warmth. The album traces is out on Spotify. ## Education - University of Toronto Mississauga: Honours BSc, Psychology - Humber College: Graduate Certificate, Advertising Account Management ## Skills - Technical solutioning: Discovery, solution design, demos, proofs of concept, RFP responses, and custom sandboxes built around each customer's stack. - Solution architecture: Mapping how AI agents, helpdesks, and ecommerce systems fit together, from data flows to edge cases. - Integrations and middleware: Middleware and API integrations that let an AI agent act on real orders, across Shopify, 3PL, returns, and subscription platforms; REST APIs, webhooks, Zapier, and MCP servers. - Development: Custom Shopify apps, demo sites that mirror a prospect's storefront, internal tools, and full-stack web apps, in TypeScript, JavaScript, and Python with Next.js and React. He designed and built GatherDwell end to end, including its AI agent, RAG pipeline, REST API, and MCP server. He also built Mailroom, a Mac email app, through to signed and self-updating releases. Most recently he built Unhallowed, a browser game on the original Diablo's rules, in a week. - Deployment and infrastructure: Vercel and Cloudflare Workers, SQL databases like Turso and Cloudflare D1, sign-in with Clerk, and continuous deployment from GitHub. - AI: AI agents, retrieval (RAG) pipelines, LLM guardrails and cost controls, and model evaluation. - Go-to-market: Enterprise and mid-market sales, pipeline generation, and building a sales function from the ground up. - Leadership and enablement: Leading Gorgias' Technical Solutions team of 8 on an interim basis, AI enablement and internal tooling across go-to-market, plus training and mentoring: SDRs, account executives, and four direct reports who all became managers. Top skills on LinkedIn: retrieval-augmented generation (RAG), the Model Context Protocol (MCP), and Next.js. LinkedIn headline: Principal Solutions Engineer and Interim SE Manager at Gorgias, AI Agents for Ecommerce CX. ## Certifications and languages - Salesforce Certified Administrator - Trailhead Double Star Ranger - Challenger Fundamentals: Core Sales Skills - Languages: English and Ukrainian, both native or bilingual ## Contact and links - Email: asavytski@gmail.com - LinkedIn: https://www.linkedin.com/in/antonsavytski - X: https://x.com/antonsavytski - Music: https://stvdwell.com - Mailroom: https://getmailroom.dev - GatherDwell: https://gatherdwell.ca - Unhallowed: https://playunhallowed.com - Website: https://antonsavytski.com - Location: Toronto, Ontario