Anass Agdi
routing ai agents in production

Anass Agdi

AI Engineer - Autonomous Systems, Automation, and Applied AI

I build production AI products: agent workflows, document pipelines, operator dashboards, and automation systems that handle real data for real users.

See What I've BuiltWork With Me
Download CV (EN)/Descargar CV (ES)
n8n Workflow • Automation Pipeline
Autonomous Workflow Pipeline
  • 3 Live Products with Demos, Auth, and Billing
  • AI Systems: Parsing, Agent Dashboards, and Automation
  • End-to-End Ownership: Build, Deploy, and Iterate

Shipping proof

Selected Works

A few representative projects that show product judgment, system design, and execution.

Founder & Principal Architectv4.0.1 - Public npm release

AgdiAgdi.dev

A local-first AI framework for running, observing, and steering multi-agent workflows.

Agdi is an engineering framework for local execution with remote visibility. It connects terminal-level workflows to browser-based operator tooling through a secure local gateway, with support for multi-agent orchestration, observability, and practical automation.

TypeScriptReactNode.jsWebContainersGemini AIPlaywrightCommander.jsVercel

Multi-Agent Execution

Five specialized agents can work in parallel, with small reversible changes instead of one opaque run.

WebSocket Local Gateway

A secure authenticated WebSocket node bridges local execution and browser tooling without hiding what happens on the machine.

Review and Audit Controls

Human checkpoints and automated QA scans make the system easier to trust in production.

17
CLI Commands
5
AI Agents
13+
LLM Models
83
Automated Tests
7
Providers
Live DemoGitHubnpm i -g agdi@4.0.1
Agdi
Agdi detail view

Architecture notes

Local execution bridged to cloud interfaces through a secure gateway.
Auditability and approval checkpoints are built into the operator flow.
The UI acts as an operator control plane, not a decorative showcase.
Agdi Dashboard - AI Agent Command Center
Live Analytics
System Health
AI Engineer & Automation ArchitectProduction - 30+ pages

Agdi Dashboard - AI Agent Command Center

A real-time operator dashboard for monitoring, configuring, and steering AI agents from the browser.

Agdi Dashboard is the browser control plane for the Agdi framework. It connects to the gateway over WebSocket and exposes fleet telemetry, traces, approvals, knowledge workflows, and agent management in a single interface.

Next.js 16TypeScriptWebSocketJSON-RPC 2.0LanceDBPlaywright

Multi-Agent Fleet

Real-time status monitoring of all AI agents with live health indicators and command dispatch.

Live Analytics

Real-time metrics make it easier to spot drift, load, and system health issues.

RAG Knowledge Base

Documents can be ingested into LanceDB for retrieval-augmented workflows and internal knowledge use.

Approval Workflows

Push notifications support human approval gates when an agent needs review before execution.

View on GitHub
DockParser

DockParser

Invoice and contract parsing with structured extraction, confidence scoring, and billing-ready workflows.

Next.jsGeminiStripe
VendorVault

VendorVault

Secure document approvals and vendor onboarding for enterprise ops.

ReactSupabase
Nano-Banana

Nano-Banana

A lightweight AI concept generator built around fast interaction, playful UX, and image output.

AIGenerative UX

Systems I've Built

Production code. Real tradeoffs. Actual results.

DockParser

AI document parser that extracts structured data from invoices and contracts.

// tradeoffs: Chose Gemini over GPT-4 for vision tasks — 3x cheaper, faster cold starts. Server-side only to avoid key exposure. Rate-limited demo mode to control costs.

→ 90% reduction in manual data entry time. Handles 50+ document formats.

Measured on 50 document test set, internal pilot.

B2B Lead Automation

End-to-end cold outreach pipeline: intake → research → personalized email → CRM sync.

// tradeoffs: n8n over Zapier — self-hosted, no per-task limits. Gemini Vision for OCR on screenshot leads. Error nodes with Telegram alerts for failed runs.

→ Processes 200+ leads/day. 15% reply rate on cold outreach.

Last 30 days of client campaigns, tracked via CRM.

VendorVault

Document approval workflow for vendor onboarding with role-based access.

// tradeoffs: Supabase RLS over custom auth — faster to ship, battle-tested security. Edge functions for webhooks. No external auth provider to reduce deps.

→ Full CRUD + approval flow shipped in 2 weeks. Zero security incidents.

Production system since Q4 2025, zero breaches reported.

Case Study: DockParser

A document-processing product built around confidence scoring, secure server-side inference, and billing-ready infrastructure.

The Challenge

Manual processing was the bottleneck.

Manual invoice processing consumes time and causes errors. Operations teams spend hours on data entry instead of strategic work.

Outcome:

About 90% less manual data entry in test scenarios, with production-ready data handling and confidence-aware extraction.

The value here is not just OCR. It is a production-safe pipeline with cost controls, confidence scoring, and billing-ready infrastructure.

Request the Technical Walkthrough

Technical Decisions

Gemini over GPT-4

3x cheaper for vision, faster cold starts

Server-side only

No client API calls — zero key exposure

Supabase RLS

Auth + storage in one, no custom RBAC code

Rate-limited demo

Controls costs, prevents abuse

System Architecture

React Frontend

Guided upload and review flow for invoices and contracts.

Edge Processing

Secure server-side inference and cost-aware request handling.

Gemini Vision

Extracts structured data and returns confidence-aware output.

Supabase

Stores documents, user data, and structured extraction results.

Stripe Billing Layer

Subscription-ready metering and webhook handling for production use.

For hiring teams

Why teams hire me for AI product work

I build AI products end to end: interfaces, backend systems, orchestration, and production constraints that hold up outside demos.

Best fit roles

AI / Product EngineerApplied AI EngineerFull-Stack Product Engineer

Strong fit for teams that need someone who can move between product thinking and implementation. I can prototype quickly, make architecture tradeoffs, and ship production-grade AI features without treating the frontend, backend, and automation layer as separate jobs.

Download CV (EN)Descargar CV (ES)Talk about a role

High ownership

I can take a problem from product framing through implementation, deployment, and iteration.

Shipping mindset

I optimize for working systems, practical tradeoffs, and measurable progress instead of speculative architecture.

Systems depth

I work across workflows, APIs, data stores, auth, operator tooling, and AI orchestration.

Product judgment

I care about user flow, clarity, and whether the system is actually useful once it leaves the prototype stage.

About Me

I like hard problems, fast iteration, and building things that hold up in production.

6 Years Calisthenics

Years of training taught me patience, consistency, and how to keep showing up.

3 Years Entrepreneur

I have spent the last few years building products, testing ideas, and learning quickly.

Discipline & Consistency

I care more about steady output and good decisions than short bursts of motivation.

Proof

@anassagd432agdi on npm3 Live Demos

Open-Source Workflows

Production-ready n8n automations — free to download and customize

B2B Lead Generation & AI Qualification

B2B Lead Generation & AI Qualification

Download JSON

End-to-end cold outreach automation. Takes raw leads → enriches → qualifies → sends personalized emails → logs everything.

  • +Lead intake from Google Sheets or Telegram
  • +Gemini Vision OCR for screenshot/photo leads
  • +AI research agent pulls context: site, services, proof points
  • +AI copywriter generates personalized cold emails
n8nGemini AIGmailTelegramGoogle Sheets
Website Contact Form → AI Analysis → CRM

Website Contact Form → AI Analysis → CRM

Download JSON

Full-stack contact form automation. Webhook receives form → Gemini extracts & qualifies lead → syncs to HubSpot → sends auto-response + follow-up → notifies Telegram.

  • +Webhook trigger captures website form submissions
  • +Gemini AI extracts business details & assigns lead score (1-10)
  • +Instant Telegram notification with AI-generated summary
  • +Auto-sync to HubSpot CRM with lifecycle stage mapping
n8nGemini AIHubSpotGmailTelegramWebhooks

More workflows coming soon. Follow me on X/Twitter for updates.

Let's Build Something Useful

Open to AI product engineering roles, selective remote contracts, and business automation projects.

Hiring path

Evaluating me for a role or contract?

Use the CV downloads and send a short note with the team, stage, and kind of product or AI systems work you need.

Download CV (EN)Descargar CV (ES)

Service path

Need help shipping automation for a business?

If you are a business owner looking for AI intake, internal tooling, or workflow automation, the service path is still open through Agdi.

Email

Hiring? Send a short brief to info@anassagdi.site

Business automation

Need a partner for AI systems work?

I still take on selected automation and AI implementation work for business owners, but this portfolio is primarily about how I build and ship product systems.

Visit Agdi AI

Secondary path for service work

AI Receptionist (24/7)

Automated call handling, lead qualification, and appointment booking. Integrates directly with Google Calendar.

AI Sales Website

High-converting websites with embedded Voice + Chat AI. Your digital sales rep that qualifies leads and schedules instantly.

Custom AI Automations

Streamline workflows with email automation, outreach sequences, and Telegram-based AI personal assistants.

Powering Production Systems

TypeScript
Next.js
Supabase
PostgreSQL
n8n
Gemini

Secure server-side ops • BYOK ready • zero client key leaks

© 2026 Anass Agdi — Agdi AI

Built and maintained by Anass Agdi.