AI · Agentic Workflows · Client Work (NDA)
AI Agents & Workflow Automation
LLM-powered agents that run multi-step workflows and remember the user across sessions — built for production, not demos.
The problem
Most LLM features stop at a single prompt and response: they can't take actions across several steps, and they forget the user the moment a session ends.
The client needed AI that does real work inside their product — calling the right tools in the right order, and responding with context about each user instead of starting from zero every time.
My approach
Agents with tools, not just prompts
Designed agents around tool/function calling so the model can fetch data and trigger actions in the application, with each tool exposed as a small, well-defined API.
Workflows as explicit steps
Broke multi-step tasks into orchestrated workflow steps, so each stage is observable, testable and easy to change without rewriting prompts.
Memory as a first-class layer
Built short-term memory for in-conversation context and long-term memory that persists user context across sessions, so responses stay personalised and relevant.
The solution
A set of production AI agents integrated into the client's application, running multi-step workflows end to end through tool calling.
A memory management layer that lets agents recall prior context across sessions, delivering consistent, context-aware responses for each user.
The outcome
2-tier
Agent memory (short + long term)
Multi-step
Agentic workflows
LLM
Tool / function calling
Next project
Placement Management System →