Skip to content
FAHD MOHAMED
All work

Professional work · Competitor intelligence · Python

Agent X

Watches competitors’ websites and news, detects real changes deterministically, and lets an LLM explain only what matters.

pytest files
37
pytest files
source: tests/
Daily digest email rendered by the project’s template

The problem

Competitor monitoring tools drown teams in noise: every cookie banner and date change looks like news.

My role

  • Engineered the system: AI agent pipeline, LLM integration and intelligent interpretation

Engineered during my role as AI & Software Engineer at Elenor Marketing Agency, Cairo (Jan 2026 – present). Screenshots come from sanitized demo environments: client and business data are replaced with demo data.

What it is

A Python pipeline fetches competitor pages, sitemaps, RSS, GDELT news and Wayback snapshots, normalizes them into semantic blocks, fingerprints them, and promotes a change only when it persists (candidate → confirmed). Changes are scored deterministically. The LLM is called only to interpret low-confidence events, in JSON mode validated by Pydantic, under a request budget. A daily digest and weekly trend report are emailed.

Key features

  • Deterministic change detection with confirmation windows
  • Budgeted, schema-validated LLM interpretation
  • Daily digest and weekly trends via Resend
  • Source-health monitoring

Architecture

Fetch

  • Pages · sitemaps
  • RSS · GDELT
  • Wayback

Detect

  • Semantic blocks
  • Fingerprints
  • Candidate → confirmed

Interpret

  • Deterministic score
  • LLM (low confidence only)

Deliver

  • Daily digest
  • Weekly trends

AILLM used only where rules are uncertain

AIPydantic-validated JSON output

AIPer-run request budget

Stack

Core

Python 3.11uvhttpxPlaywrightBeautifulSoupfeedparser

Data

PostgreSQL (psycopg 3)Pydantic 2

AI & mail

OpenRouterJinja2Resend

Quality

pytest (37 test files)

Engineering challenges

01

Spend the model where it counts

Most changes are classified by rules. The LLM sees only ambiguous ones, which keeps cost bounded and every insight traceable to before/after evidence.

The digest is the project’s real render_digest() output fed with fictional data. The full pipeline was not run (it needs OpenRouter and Resend keys).

Next project

Church QR Attendance