AI Automation Developer  ·  Voice · SMS · LinkedIn · Email · RAG
Open to work — remote, worldwide

Sovon Shaik

I build AI systems that call — while you sleep.

Voice agents, SMS agents, LinkedIn agents, RAG knowledge agents and end-to-end outbound engines built on n8n, Supabase, Retell, Vapi and custom API layers. Self-hosted infrastructure keeps the cost of running all of it close to zero. Not demos — production systems that run unattended, every day, at scale.

See the systems → Book a call ✉ sovon@aimamoth.com aimamoth.com
24/7Unattended operation
5 channelsVoice · SMS · Email · LinkedIn · RAG chat
Self-hostedn8n · scraper · email finder
Open to workRemote · full-time or contract
AI Voice AgentsAI SMS AgentsCustom LinkedIn AgentRAG Agentsn8n OrchestrationSelf-Hosted InfraSupabaseRetell AIVapiElevenLabs Voice CloningComposioLocal OllamaGoHighLevelZoho CRMKEAP CRMAirtableCustom API IntegrationsLead GenerationCold Email Infrastructure AI Voice AgentsAI SMS AgentsCustom LinkedIn AgentRAG Agentsn8n OrchestrationSelf-Hosted InfraSupabaseRetell AIVapiElevenLabs Voice CloningComposioLocal OllamaGoHighLevelZoho CRMKEAP CRMAirtableCustom API IntegrationsLead GenerationCold Email Infrastructure
Actively building — updated August 2026
Right now

What I'm building this month

Not a portfolio that stopped shipping the day it went live — here's what's on my bench right now.

AEO Optimization for Doctors

Getting medical practices found and cited by AI answer engines — ChatGPT, Perplexity, AI Overviews — not just ranked on Google. Structured content, schema and citation-worthy pages built for how patients search now.

Autonomous Lead Pipeline — SMS + Voice

An end-to-end pipeline that sources, qualifies and books leads across SMS and voice with zero manual touch — the next iteration of the systems already running for clients.

Testing Meta Ads with Claude

Running Meta ad creative, targeting and copy experiments with Claude in the loop — generation through performance analysis — to see how far AI can take ad ops.

Launching My Own SaaS

Turning the infrastructure I've built for clients into a product of its own. More to share soon.

By the numbers

Real output, from a real client

These figures come from live campaigns run for a small US business — not a lab test. Modest volume, but every meeting below was sourced, qualified and booked without a human touching it.

12
Qualified meetings booked autonomously
voice agent + email engine
9
Qualified meetings from the AI voice agent
Project 01 · booked into GHL mid-call
3
Meetings booked from cold email
Project 02 · fully automated sourcing
2–4%
Cold email reply rate
Project 02 · at or above industry norm
4%
SMS response rate
Project 03 · self-built infrastructure
10–15%
LinkedIn connection acceptance
Project 04 · cold, no warm intro
4
Outbound channels orchestrated
voice · SMS · email · LinkedIn
24/7
Unattended operation
retries · fallbacks · alerting

Small numbers, honestly reported. The point isn't the volume — it's that the entire path from cold list to booked calendar slot ran unattended, and every one of these systems is architected to scale on the same rails.

What I do

Autonomous systems, not automations

Most "automation" still needs a human babysitting it. I design the retry logic, state handling and escalation paths so the system keeps running when things go wrong — which is the only thing that matters in production.

AI Voice Agents

Outbound and inbound conversational agents on Retell and Vapi — shipped for real estate, SEO agencies, marketing teams and more. Call scheduling, retry cadences, live qualification, objection handling, calendar booking mid-call, and deep custom API integration into whatever CRM or dashboard the client already runs.

AI SMS Agents

Two-way SMS conversation engines with intent detection, human handoff and message routing into whatever inbox the operator already lives in.

Custom LinkedIn Agent

A self-built LinkedIn outreach agent — prospect discovery, connection requests, AI-written personalised follow-ups and reply detection, all sequenced with human-like pacing and daily limits.

Workflow Orchestration

n8n as the nervous system — webhooks, queues, conditional branching, error workflows and scheduled triggers wiring every tool into one pipeline.

Data & Backend

Supabase and Airtable as the system of record — lead state, call outcomes, dedupe logic, and API endpoints custom-built where off-the-shelf tools stop.

Outbound at Scale

Cold email infrastructure, deliverability hygiene and AI-personalised copy per prospect, fed by lead generation across Apollo, Icypeas, Prospeo and a self-built Google Maps scraper — enrichment running in parallel across every channel.

Self-Hosted Infrastructure

My own self-hosted n8n instance, a self-hosted website scraper and an email-finder automation running inside VS Code — a stack I own end-to-end, which keeps infra cost close to zero. Local Ollama models handle a lot of the day-to-day work that doesn't need a hosted API.

AI Content & Social Automation

Composio-driven autoposting across LinkedIn and Instagram, video and avatar generation with HeyGen and Higgsfield, and Claude Code as a daily driver — including cloning useful GitHub repos and turning them into custom Claude skills.

Selected work

Systems running hands-off

Each of these was built end-to-end — architecture, integrations, agent design and the failure handling that keeps them alive.

PROJECT 01 — REAL ESTATE

AI Voice Agent for Distressed Property Outreach

Voice agents · n8n · Airtable · GoHighLevel
Fully autonomous — zero human touch

An AI voice agent that calls distressed property owners three times a day on a managed cadence, holds a real qualification conversation, and books the meeting straight into GoHighLevel while the prospect is still on the line. The owner gets an instant SMS notification the moment a booking lands.

Airtable list→ 3× daily call trigger→ AI voice qualification→ GHL booking→ SMS alert
9Qualified meetings booked
3×/dayCall cadence per lead
0Human hours per booking
AI Voice Agentsn8nAirtableGoHighLevelSMS
PROJECT 02 — OUTBOUND ENGINE

Indeed-to-Inbox: Autopilot Cold Email Machine

Indeed API · Enrichment · AI copy · Email infrastructure
Runs on autopilot — no manual step

Scrapes live job listings via the Indeed API, resolves each listing to a company, identifies the right decision-maker, finds and verifies their email, then generates a genuinely personalised message per contact and sends at scale. Hiring signals become booked conversations without anyone opening a spreadsheet.

Indeed API→ Company resolve→ Decision-maker→ Email find + verify→ AI personalisation→ Send at scale
2–4%Reply rate
3Meetings booked
100%Sourcing automated
Indeed APIn8nEmail VerificationAI PersonalisationCold Email
PROJECT 03 — INFRASTRUCTURE

Custom SMS AI Agent with Live Dashboard + Telegram Control

Custom API layer · Telnyx · Twilio · Telegram · Dashboard
Built from scratch — own the whole stack

Rather than paying for a rigid SMS platform, I built my own — complete with a live dashboard to track conversations, response rates and lead status at a glance. Messages go out to US prospects through a custom API layer over Telnyx and Twilio, and every reply also lands in Telegram — so the whole conversation can be handled from a phone, in a thread, with no CRM tab open. Full control over sending logic, threading and routing.

Dashboard↔ Telegram thread↔ Custom API layer↔ Telnyx / Twilio↔ US prospect
4%Response rate
<1 minReply-to-Telegram latency
1Live ops dashboard
$0Third-party platform fees
Custom APITelnyxTwilioTelegram Bot APIDashboardn8nSupabase
PROJECT 04 — SOCIAL OUTBOUND

Custom LinkedIn Agent — 24/7 Two-Way Conversations

Custom API layer · ElevenLabs · Composio · n8n · Supabase
Self-built — not a sequence tool, a conversation partner

A LinkedIn agent I built rather than rented — and one that does more than fire a sequence. It sources prospects against an ICP, opens the connection with a cloned ElevenLabs voice message as the first touch, follows up with images and AI-personalised text, then automatically replies to every incoming message, 24/7. Most tools on the market can only run a one-way drip; mine holds an actual back-and-forth conversation around the clock and only hands off to me once a lead is genuinely interested. I run the same Composio integration to autopost content across LinkedIn and Instagram.

ICP prospecting→ Connection request→ ElevenLabs voice note→ Image + AI text→ 24/7 auto-reply→ Handoff
10–15%Connection acceptance
24/7Auto-replies to inbound
3Formats — voice, image & text
0Account restrictions
LinkedIn OutreachCustom APIElevenLabs Voice CloningComposioAuto-ReplyAI Personalisationn8nSupabase
PROJECT 05 — KNOWLEDGE & CONTENT

Book → RAG Agent + Content Marketing Pipeline

RAG · Vector embeddings · Local Ollama · n8n · Content automation
One knowledge base, two products

I took a full book, chunked and embedded it into a vector store, and wrapped it in a RAG agent that answers questions and holds a conversation grounded entirely in that book's material — no hallucinated citations, just retrieval against the real text. The same knowledge base doubles as fuel for an automated content marketing pipeline: it extracts ideas from the source material, drafts on-brand posts with AI, and repurposes them across channels on a schedule, no writer's block involved.

Book ingestion→ Chunk & embed→ Vector store→ RAG chat agent + Content pipeline→ Scheduled posts
1Book → full knowledge base
24/7Conversational RAG agent
Multi-channelContent pipeline output
RAGVector EmbeddingsLocal Ollaman8nSupabaseContent Automation

…and a long tail of systems built on n8n

These five are the highlights. The rest — internal ops bots, CRM syncs, AI research pipelines, reporting automations, custom clinic dashboards, scraping infrastructure — live at aimamoth.com.

CRM sync pipelines AI research agents Google Maps scraper Custom clinic dashboards Reporting automations Internal ops bots Webhook middleware Email finder automation
Explore more at aimamoth.com →
Tech stack

What I build with

Tool-agnostic by preference — I pick whatever gets the system to production fastest and keeps it maintainable afterwards.

AI & Voice

Retell AIVapiElevenLabs Voice CloningOpenAIAnthropic ClaudeLocal OllamaPrompt engineeringAgent design

Automation & Orchestration

n8n (self-hosted)MakeComposioWebhooksCron / schedulersError workflowsQueues

Data & Backend

SupabasePostgreSQLAirtableGoogle SheetsREST APIsCustom API integrationsCustom dashboards

CRM & GTM

GoHighLevelHubSpotZoho CRMKEAP CRMInstantlyTelnyxTwilioTelegram Bot APICalendar APIs

Lead Gen & Enrichment

ApolloIcypeasProspeoOwn Google Maps scraperEmail finder automationList building & enrichment

AI Content & Creative Tools

Claude CodeHeyGenHiggsfieldComposio autopostingGitHub repos → Claude skills

Self-Hosted Infra

Self-hosted n8nSelf-hosted web scraperEmail finder (VS Code)Local Ollama modelsMinimal infra cost by design

Growth Skills

Lead generationEmail marketingDeliverabilityLinkedIn outreachOutbound copywritingRAG & content marketing

Engineering Basics

JavaScriptPythonJSON / data mappingAuth & API keysRate limitingLogging & monitoring
How I work

From problem to unattended system

The build is the easy part. Making it survive real-world edge cases is where the work actually is.

Map the process

Understand the manual workflow first — the exceptions, the judgement calls, the bits nobody documented.

Design the architecture

Decide what holds state, what triggers what, and where the human still needs to be in the loop.

Build & integrate

Agents, workflows, APIs and database wired together — with retries and fallbacks built in from day one.

Harden & hand off

Monitoring, error alerting and documentation so the system runs without me standing next to it.

The unglamorous part

Why these don't break

Anyone can wire a happy path. The difference between a demo and a system someone trusts with their pipeline is what happens on the bad days.

Retries & idempotency

Every external call has a retry policy and a dedupe key. A timeout on a provider doesn't double-send an email or double-book a calendar slot.

Error workflows & alerting

Dedicated n8n error workflows catch failed executions and push a readable alert with the payload — so a break is noticed in minutes, not at month-end.

Rate limits & compliance

Human-paced sending windows, per-account daily caps, suppression lists and opt-out handling built in — so volume never costs you the account or the domain.

Documentation & handoff

Architecture notes, env/credential inventory and a runbook per system — so the business isn't hostage to the person who built it. Including me.

If you hire me

What the first 30 days look like

No ramp-up theatre. Here's the plan I'd run from day one.

Week 1 — Audit

Inventory every existing workflow, integration and manual process. Find what's silently broken and what's costing the most hours.

Week 2 — Quick win

Ship one automation that removes a real, measurable chunk of manual work. Build trust with output, not with a roadmap deck.

Week 3 — Harden

Add monitoring and error handling to the systems already running. Stop the silent failures nobody's tracking yet.

Week 4 — Scale plan

Propose the next three builds ranked by hours saved and revenue touched, with honest effort estimates attached.

Let's talk

Ready to build systems that run themselves

If you're hiring an AI Automation Developer who can own a system end-to-end — architecture, agents, integrations and the boring reliability work — I'd like to hear about the problem you're solving.

📅 Book a call with me → sovon@aimamoth.com
Available for full-time or contract Remote · comfortable across AU, US & EU hours Response within 24 hours