← Back to work

Case Study · 2025

Sales Detect Agent

An autonomous multi-agent system that watches e-commerce deals, estimates competitive prices, and fires real-time alerts.

RoleSoftware Engineer
Timeline2025
FocusLangChain · LLM · Agentic AI
SDA

Context

E-commerce deals move faster than a human can re-read them. The system had to discover, compare, and alert — without spamming.

The Problem

Manual price watching is late and noisy. A single agent mixing discovery, analysis, and notification gets all three wrong.

Process

01

Split the jobs

Specialized agents for product discovery, price analysis, and notification orchestration — each with an LLM and a clear contract.

02

State and freshness

Intelligent scheduling and state management to track deal freshness and prevent alert fatigue.

Decisions

Many agents, one coordinator

One giant prompt hallucinates prices. Narrow agents correct each other.

Silence is a feature

Without freshness tracking, every deal re-fires. State stops the spam.

Outcome

Autonomous real-time alerts on e-commerce deals

Less noise — deal freshness is checked before notify

Reflection

An agent is only useful if it knows when to stay quiet. Orchestration is as much filtering as action.