AI Agents & Workflow Development
Systems that read and understand information, apply business rules, make recommendations, call external tools, and complete multi-step tasks — connected to the places your work actually happens.
I'm Sanjaya Jayathilaka, a developer from Sri Lanka. I take slow, repetitive, manual processes and rebuild them as reliable systems — AI agents, workflow automation, web scraping, and custom backends.
Most of my work starts with a process that is slow, repetitive, or hard to manage by hand.
I study how that process actually works, identify which decisions can be automated, and build a system that handles the work reliably. My main focus now is AI agent and workflow development — systems that read and understand information, apply business rules, make recommendations, interact with external tools, and complete multi-step tasks end to end.
My background in web scraping makes the AI work sharper, because an agent is only as useful as the information it can reach. I've spent years pulling structured data out of dynamic sites, login-protected platforms, directories, and other difficult sources — then cleaning, matching, deduplicating, and shipping it somewhere useful.
I choose tools according to the project instead of forcing every problem into the same framework. Sometimes AI is the centre of the solution. Sometimes a plain algorithm, an API integration, or an automation script is simply the better answer.
My job is to know the difference.
Understand the real business problem — then write code.
Inputs, outputs, decision rules, edge cases, success criteria. Clarified before anything gets built.
Five disciplines that share the same backbone — backend logic, API communication, data processing, and systems that run without constant supervision.
Systems that read and understand information, apply business rules, make recommendations, call external tools, and complete multi-step tasks — connected to the places your work actually happens.
Multi-step processes wired together across n8n, APIs, email, Sheets, CRMs, forms, and messaging platforms — with real error handling, not happy-path demos.
Extraction from dynamic sites, login-protected platforms, directories, and e-commerce sources — then cleaned, matched, deduplicated, and exported wherever you need it.
APIs, databases, integrations, and the unglamorous plumbing that keeps everything running — built to be maintained, not just demonstrated.
Slash commands, moderation tools, database systems, API integrations, interactive buttons and menus, and real-time data features.
Agents and scripts that drive real browsers — filling forms, navigating portals, and handling the web tasks nobody should do by hand twice.
Depending on the project, an agent or workflow might look like any of these.
Resolves common inquiries, applies business rules, and routes edge cases to human teams.
Evaluates incoming leads against criteria, collects key details, and updates the CRM automatically.
Extracts structured data from PDFs, forms, and invoices for automated downstream handling.
Compares listings and inventory requirements to deliver instant, rule-based recommendations.
Handles booking availability, sends automated confirmations, and syncs calendar events.
Scrapes dynamic and login-protected sources, handling anti-bot tools, proxies, and cleaning.
Queries internal documentation and vector stores so team members get instant, verified answers.
Triggers targeted follow-ups, syncs contact updates, and maintains multi-channel messaging pipelines.
Chosen per project, never forced. The right combination beats the familiar one.
I don't start by writing code. I start by understanding the real business problem.
Every system I ship is observable — you can see what it decided and why.
Before any code: what are the inputs? The expected outputs? The decision rules, the edge cases, and what "working" actually means for you.
Decide which parts genuinely need AI and which are better served by a plain algorithm, an API integration, or a straightforward script. Then pick the tools to match.
For larger projects I build a working minimum version first and demonstrate it early. You see real behaviour on real data long before the budget is spent.
Collect feedback, improve in defined increments, harden the error handling, and hand over something that keeps running without constant manual attention.
Tell me about the process you're tired of doing by hand. I'll tell you whether it's worth automating — and how.
A few short questions about the process you want to fix, what you've already tried, and your timeline. I read every submission myself.
Opens a short form — no account needed.