Remote access & execution
A.R.I.S. connects a cloud assistant to my physical execution node. I can send a request through Telegram or the web, and supervise the work from another device.
PERSONAL PORTFOLIO / 2026
Alex
Limin.
Software engineer. Founder of SayMore.
Lead engineer on Komunity AI.
These are some of my projects.
SELECTED WORK / 2026SELECTED PROJECTS ↓A.R.I.S. / PROJECT NOTES
Project data
September 2026
225,243market observations
142fleet units
74posts marked published
From production data07 SEP 2026 ↗
01 / A.R.I.S.
A.R.I.S.
Personal AI infrastructure.
A.R.I.S. · Personal AI infrastructure
A.R.I.S. / SYSTEM MAP01
ANYWHEREYOUWeb · Telegram · Voice
request
CLOUDFLAREA.R.I.S.Brain → job relay → D1
outbound polling
REMOTE HOSTTHE NODEUbuntu · Claude Code · Tailscale
Waiting for the next chapter…_
Architecture visualization · built from the actual services
Supervised coding sessions
The node polls for jobs, runs Claude Code, and sends back progress. When a supervised job needs an answer, it pauses and resumes the same session after my input.
Redacted events reach the dashboard. Failed deliveries stay in a local spool for retry. Each job retains its event history.
32 GBnode memory¹
60 sectelemetry interval
Automated market research
Market Scout combines SEC filings and daily market data into scheduled stock and crypto research. Deterministic checks reject or quarantine candidates; missing evidence stays unknown.
225,243ticker-day observations
13,229distinct tickers
Validation fixes +
Counting mentions of reverse splits falsely flagged Apple. I anchored the check to actual 8-K filing events instead. Another fix removed a volume-based storage floor that was quietly biasing the baseline toward busy days.
02 / CONEXIÓN COMMAND
Conexión Command
Fleet operations.
Conexión Command · Transport operations
CONEXIÓN / OPERATIONS02
FLEET UNITS142 132 reporting in 24h
8,688 ROUTE RUNSSAMSARA → WORKER → D1
Illustrated world · actual fleet totals, captured 7 Sep 2026
Dashboard & WhatsApp assistant
Conexión brings fleet observations, arrival analysis, fuel investigations, and alerts into one control tower. Operators can reach the system through its dashboard and WhatsApp interface.
Reducing database reads
The dashboard heartbeat was recounting GPS history. I replaced that repeated scan with a maintained summary.
BEFORE / ROWS READ~6,000,000
↓AFTER / ROWS READ1
Historical production measurement recorded in the build log. Query cost reduction; not a claim about response-time improvement.
Architecture +
Samsara → scheduled Worker → D1 history + KV live state. The WhatsApp webhook queues inbound events; a Worker consumer handles the assistant flow. Arrival, prediction, fuel, and alert engines expose their results to both interfaces.
03 / KOMUNITY AI
Komunity AI
Creator relationships.
Komunity AI · Creator relationship platform · Lead engineer, ViralMe
KOMUNITY / CONCEPT VIEW03
CREATOR DIRECTORY / 014Shared identity
C
Creator 014
Video creator · Lifestyle & design
InstagramTikTokYouTube
BRAND RELATIONSHIPS Private to each workspace
A
Partner AActive collaboration
PartnerCURRENT WORK
Product launch series
NEXT STEP
Review first draft
PRIVATE NOTE
Prefers a written brief before production.
B
Partner BIndependent relationship record
EvaluatingWORKSPACE / WORKExample project
Launch collection
Briefs, deliverables, and approvals
Briefed 2
Product introduction
Creator 014
Brief attachedCPartner A
Behind the scenes
Creator 027
3 framesIn progress 1
How it works
Creator 008
Draft in progressCPartner A
Review 1
First impressions
Creator 019
Awaiting approvalCPartner A
Custom portfolio render · illustrative records
Creator records & work tracking
Creator profiles, conversations, opportunities, briefs, deliverables, and approvals belong together. Komunity brings them into a workspace built around the relationship.
Shared identities, private records
A creator has one shared identity. Each brand keeps its own relationship record, private notes, and history. That separation is part of the data model.
BRAND A
Private relationship↔CREATOR
Shared identity↔BRAND B
Private relationship
Private relationship↔CREATOR
Shared identity↔BRAND B
Private relationship
Under the interface +
Web application → API Worker → Hyperdrive → PostgreSQL, with Supabase authentication. Conversation Durable Objects, R2 attachments, and separate notification and provider-sync queues handle supporting workflows.
04 / JOINBRANDS
JoinBrands
Content automation.
JoinBrands · Content automation · SayMore Inc.
JOINBRANDS / FINISHED OUTPUT04
01 / 04 EXTRACT
Original 2× output · post 81 · four slides from a six-slide carousel
The production pipeline
Source ingestion feeds idea generation. A durable workflow writes the script, renders the slides, saves them to R2, and routes the result for review or autonomous publishing.
19,277source items ingested
74posts marked published
Slack review & publishing
In review mode, a finished draft goes to Slack before publishing through Blotato. The product also includes a theme editor and multi-account administration.
Source→Script→Render→Review→Publish
What the numbers mean +
74 posts are marked published in the application database, with recorded dates from May 12 to July 31, 2026. That is not an independent count of currently visible Instagram posts. Approval decisions were not populated, so no approval-rate claim is made.
Slides are historical generated editorial content, preserved as output examples.
05 / OTHER WORK
Other
projects.
AJoinBrands support automationKnowledge → draft → human review+
A poller watches the Skool community; a reasoning service uses a Vectorize knowledge base; an approver routes drafts through Slack. A separate project from the content pipeline above. Deployment status is owner-reported in the supplied portfolio.
BIntercom audit & replacement planRequirements · spend analysis · buy / build / hybrid+
Requirements interviews, a twelve-month spend breakdown, and a costed recommendation turned an ambiguous business question into a decision. Analysis work, listed as in progress in the supplied portfolio.
CCreator intelligenceResearch for partnership targeting+
A scraped and classified view of creators, delivered as a workbook for partnership targeting. The source workbook is still pending, so no record counts or outcome metrics are claimed.
ABOUT
L_Leonid Alexander
Limin.
I go by Alex. I study Political Science at UBC and work on software, automation, and AI systems.
At SayMore, I build automation and AI systems for businesses. At ViralMe, I lead engineering on Komunity AI. My work covers backend services, product interfaces, integrations, and production operations.
TypeScriptPythonSQLCloudflareAI systemsEN / ES / RU