Ishoka
Robert Tega
I help startups and scale-ups replace manual bottlenecks withself-running automation ecosystemsthat operate 24/7.
Ishoka Robert
Engineering Freedom
via Automation
06
Multi-Stage Case Studies
23+
Platform API Integrations
24/7
Autonomous Logic Execution
95%+
Data Extraction Accuracy
The Automator's Toolkit
I leverage a powerful stack of AI and automation tools to build resilient, self-running systems that save hundreds of hours every month.
From Code
to Cognition.
I build systems that think, act, and evolve.
I'm an AI Automation Engineer with a background in software engineering. I specialize in designing and deploying intelligent workflow ecosystems that replace manual processes with industrial-grade automation.
My work spans document processing, ETL pipelines, and AI-driven outreach. I build the infrastructure that allows businesses to scale without adding operational friction.
Engineered Stack
Precision instruments for the job.
Architecture Ethos
Decoupled & Fault-Tolerant
Case Studies
& Real Results
From 10 hours of manual work to 30 seconds of pure automation. Here is how I make it happen. ✨
Invoice Inbox Automation
Business Impact
- 10-min manual task → under 30 seconds
- Zero duplicate entries
- 100% data capture across all document types
The Problem
Admins manually opened emails, read PDFs, and typed invoice data into spreadsheets. Scanned documents were unreadable by standard tools, causing transcription errors. There was no safeguard against duplicate payments.
The Solution
A multi-stage Make.com workflow that batch-processes invoices from Gmail every 2 hours. It uses GPT-4o Vision to extract data from any document type — including scanned images — then runs a dual-layer duplicate check (by Message ID and by Vendor + Invoice Number) before logging to Google Sheets.
Executive Performance Dashboard
Business Impact
- 4 hours → under 2 minutes
- Fully deterministic accuracy
- Self-healing with automatic alerts
The Problem
Department heads spent 4+ hours manually cleaning and collating data for year-end reviews. There was no audit trail, no real-time visibility, and errors were only discovered after the board had already seen the report.
The Solution
A Make.com ETL pipeline with isolated, parallel extraction modules that pull departmental data simultaneously. A JavaScript engine handles deterministic currency and date normalisation, results are upserted to Airtable using Year-Month composite keys, and a 10-minute concurrency lock prevents data corruption. Errors trigger real-time email alerts.
Proposal Generation & Delivery
Business Impact
- 2–4 hours → under 5 minutes
- Full audit trail on every proposal
- Consistent branded output every time
The Problem
Sales reps wrote proposals manually, chased managers on Slack for approval, and reformatted documents for each client. No duplicate checks, no template consistency, no tracking once a proposal was sent.
The Solution
A decoupled n8n micro-workflow that takes raw sales notes, expands them into professional proposal sections via OpenAI, auto-populates a branded Google Doc, routes it to a manager for approval via Discord, then exports a locked PDF and emails it directly to the client. Every stage is logged in Airtable. A 30-minute watchdog detects and flags any stalled records.
Content Generation & Publishing Platform
Business Impact
- 1 idea → 3 platform-ready variations with visuals
- Reviewed and distributed in minutes
- Zero designer dependency for standard posts
The Problem
Content managers had to write separate versions of every post for LinkedIn, X, and email — each with different formats, character limits, and tone rules. Visuals required a designer. No approval workflow existed.
The Solution
A GCP-hosted Next.js platform orchestrated via n8n. A contextual scraping module pulls source data via Jina.ai, a rules injection module fetches SEO and platform guidelines from Airtable, and an AI generation module uses GPT-4o to produce up to 3 strategic content variations. DALL-E 3 handles automated visual asset generation. A role-gated approval queue controls final distribution.
B2B Lead Generation & Outreach Engine
Business Impact
- 95% reduction in time from search to outreach-ready
- 90%+ email deliverability
- Zero duplicate outreach
The Problem
Sales teams sourced leads manually from LinkedIn, copy-pasted contacts into spreadsheets, verified emails one by one, and sent the same generic message to every prospect. Slow, error-prone, and poor conversion.
The Solution
A modular n8n workflow that scrapes LinkedIn and Apollo via Apify in rate-safe batches, verifies email deliverability through Bouncer, and cross-references against an existing company registry to eliminate duplicates. A personalisation module uses OpenAI to generate a tailored 4-step outreach sequence — 3 emails and 1 LinkedIn message — per verified lead.
Voice-Based Customer Support Agent
Business Impact
- 80% Automation of First-Line Support
- 24/7 Availability
- Zero-Hallucination Guardrails
- Improved Agent Workflow
The Problem
RelayPay, a B2B SaaS company facilitating cross-border payments in Africa, faced significant friction in scaling its customer support. As their SME and startup user base grew, the support team was overwhelmed by a high volume of repetitive queries concerning onboarding, pricing tiers, and payout timelines. Support agents were losing hours daily manually searching through internal documentation and past tickets to provide consistent answers. Furthermore, the sensitive nature of international finance meant that misinformation could lead to a loss of trust, yet human agents were prone to fatigue-driven errors during peak volume.
The Solution
A production-ready, low-latency voice AI support agent built for RelayPay, leveraging Vapi and Deepgram's Nova-3 model for high-accuracy, fintech-optimised speech recognition. Powered by a RAG pipeline using n8n and Claude, the agent delivers strictly grounded responses from RelayPay's official Notion knowledge base — eliminating hallucinations around fees and compliance. An intelligent escalation engine handles edge cases with confidence, routing ambiguous queries to human agents or declining gracefully rather than guessing.