DEV Community: Djaouad Frih

How I Turn "Pay After Demo" n8n Leads Into $2-5K AI Automation Contracts

Djaouad Frih — Mon, 31 Aug 2026 23:26:07 +0000

Most freelancers run from "pay after you see it work" deals. I run toward them.

This week a lead posted on n8n community: "Fix or build a working workflow or you don't pay. $150 fixed. You pay after you see it run in your instance."

Here's why this is actually a $2-5K opportunity disguised as a $150 gig:

The Psychology

They have budget approved (they named a number)
They have urgency (explicit failure path requirement)
They've been burned before ("not a fit: unpaid tests, 31-agent builds, full agency ops")

The Upsell Path

Deliver the $150 workflow — single trigger, single destination, explicit error handling
Demo it live — they see it run in THEIR instance
Ask the magic question: "What breaks when this scales to 10 workflows?"
Propose the retainer: "I'll build your automation foundation — $2-5K/mo for ongoing workflows, monitoring, and new integrations"

My Template Response

"Hi — I saw your post. I build exactly this: n8n workflows that survive production. Happy to deliver the first one on your terms ($150, pay after demo). If it works, we can talk about the 10-workflow foundation your team actually needs. Calendly: calendly.com/djaouad"

The Numbers

1 in 3 "pay after demo" → retainer conversation
Average retainer: $3.5K/mo
Time to close: 2 weeks from demo
Zero unpaid discovery calls

The Lesson

Low-ticket, high-trust offers are the best funnel for high-ticket automation work. Stop pitching $10K projects cold. Start delivering $150 wins that prove you can handle the $10K mess.

Currently have 3 qualified leads this week (2 from Reddit, 1 from n8n). Pipeline: $18K/mo potential. Building in public so you can copy the playbook.

AIAutomation #n8n #FreelanceStrategy #AgencyGrowth #BuildInPublic

The AI Agent vs Chatbot Gap: Where Does It Actually Work?

Djaouad Frih — Sat, 29 Aug 2026 17:41:03 +0000

Unpopular opinion: 80% of "AI agents" being sold right now are just chatbots with better marketing.

A chatbot answers a question. An agent owns a workflow end-to-end — it books, qualifies, routes, updates systems, and knows when to hand off to a human.

Too many demos impress at the keynote and fall apart on the first real edge case.

I build agents that genuinely do work, and the honest truth is: the hard 90% is not the LLM. It's the grounding, the retrieval, the guardrails, the handoff — every piece of boring infrastructure around the model. That's what separates a product from a party trick.

A question for founders, operators and engineers:

👉 Where have you seen AI actually do work in production — not in a demo? And where did it quietly fall apart?

Best and worst experiences welcome. Real ones only.

Case Study: Turning Missed Calls into Booked Leads With an AI Receptionist

Djaouad Frih — Sat, 29 Aug 2026 17:25:57 +0000

The problem

A busy service business — clinic, salon, law firm — gets calls during the exact hours nobody's at the front desk. Every missed call is a lead that books somewhere else. Staff spend their day interrupting real work to answer the same five questions, then chase leads manually.

The fix

A 24/7 AI receptionist, trained on the business's own data. It doesn't pretend to be human. It does three things that directly move revenue:

Never misses a lead — answers every inbound message in under a second, even at 2am
Books, qualifies, routes — filters serious leads, books the slot into the calendar, routes to the right department
Hands off with context — escalates to a human with an AI-drafted reply, so nothing is lost

How it ticks

The agent is grounded in a RAG knowledge base built from the business's own documents — services, prices, policies, FAQs. Everything it says is cited to real data, so it never invents prices or promises.

Multi-tenant: every business gets an isolated knowledge base
Streaming: answers start in <1 second; customers don't bounce
Native + web: one React Native codebase, plus an embeddable widget for any site or WhatsApp
Human handoff: the person gets a drafted reply with full context

The stack

Layer	Choice
App (web)	Next.js
App (mobile)	React Native
Backend	NestJS
Vector store	PostgreSQL + pgvector
Realtime	Socket.io
LLM	Gemini

What "done" looks like

A business gains an always-on front desk that answers, books, and qualifies — and the human team only steps in where judgement actually matters. That's the difference between "AI demo" and "AI that does work."

See it live

This isn't a mockup — the agent is running in production on my own portfolio. Ask it about pricing, availability, or drop it a project brief:

→ Try the live agent: chat.djaouad.tech

Two more systems I run in production:

Ask-your-anything documents (cited answers over PDFs): docs.djaouad.tech
AI customer support (order checks, tickets, escalation): customer.djaouad.tech

If you want this applied to your business: hire an AI developer — fixed price, milestone-based.

I Built a Self-Improving AI Receptionist That Books Clients — Here's the Stack

Djaouad Frih — Sat, 29 Aug 2026 17:24:46 +0000

A 24/7 AI receptionist that answers in under a second, books appointments, captures and qualifies leads, and hands off to humans with AI-drafted replies — that's what I shipped to production. Here's exactly how it's built.

The problem

Most small businesses (clinics, salons, law firms) lose leads when no one answers the phone. This agent never misses one: it books the slot, sends reminders, and only escalates when it genuinely should.

The stack

Frontend/app: Next.js + React Native (one codebase, web + Android)
Backend: NestJS
Vector store: PostgreSQL + pgvector
Realtime: Socket.io (streaming chat)
LLM: Gemini

The RAG knowledge base

The agent is grounded in the business's own data — services, prices, FAQs, policies. Documents are auto-extracted, chunked, and embedded into pgvector. When a customer asks, we retrieve the relevant chunks and the model answers with citations. No hallucinations, no inventing prices.

// chunk → embed → query pipeline (simplified)
const chunks = await chunkDocument(pdfText)
const embeddings = await embed(chunks)
await insertVectors(tenantId, chunks, embeddings) // pgvector
// query time
const q = await embed(userQuestion)
const top = await similaritySearch(q, tenantId)
return llmAnswer(top, userQuestion)

Why it works

Multi-tenant: every business gets its own isolated knowledge base
Streaming: answers start in under a second — customers don't leave
Human handoff: it writes the draft reply so staff with the context
Embeddable widget: drops into any website or WhatsApp

Try it

The live agent is running on my portfolio — ask it anything about my work, pricing, or availability:

→ Live demo: chat.djaouad.tech

I also built a Smart PDF Workspace (ask-your-documents with citations) and an AI Customer Support Agent — all in production.

If you're building AI agents or need one for your business: hire an AI developer.

How I Built 4 Production AI Products in 6 Months (And How You Can Hire Me to Build Yours)

Djaouad Frih — Thu, 27 Aug 2026 20:23:46 +0000

I'm Djaouad Frih, a Full-Stack AI Engineer. Over the last 6 months I've shipped 4 production AI products that are live, testable, and solving real problems right now. Here's what I built, how I built it, and what I can build for you.

The 4 Live Products

1. AI Virtual Receptionist — `chat.djaouad.tech`

Problem: Clinics, salons, and service businesses miss calls during busy hours. Those callers book elsewhere.
Solution: A 24/7 receptionist that answers in <1 second, books appointments into your calendar, captures leads, and hands off to humans when needed.
Stack: Next.js + NestJS + PostgreSQL/pgvector + WebSocket for real-time
Live demo: chat.djaouad.tech

2. Smart PDF Workspace — `docs.djaouad.tech`

Problem: Teams waste hours searching through contracts, policies, and technical docs for specific clauses.
Solution: Drop any PDF (even 40+ pages), ask a question, get a cited answer with page numbers. Multi-tenant, embeddable as a widget.
Stack: Next.js + NestJS + pgvector + document chunking/embedding pipeline
Live demo: docs.djaouad.tech

3. AI Customer Support Agent — `customer.djaouad.tech`

Problem: Support teams answer the same questions repeatedly. Customers wait hours for simple answers.
Solution: Reads your documentation once, then resolves tickets, checks order status, and escalates with full context. Built with LangGraph for multi-step reasoning.
Stack: Next.js + NestJS + LangGraph + WebSocket + tool-calling
Live demo: customer.djaouad.tech

4. HireMe MCP Server — `mcp.djaouad.tech`

Problem: Clients want to vet freelancers but portfolios are static and hard to query.
Solution: The first portfolio an AI assistant can read and hire from directly. Exposes my work, pricing, and availability as MCP tools — Claude/Cursor can query my profile and file a project brief automatically.
Stack: MCP protocol + TypeScript + my live product data
Live demo: mcp.djaouad.tech

The Common Architecture

All 4 products share a production-grade foundation:

┌─────────────────────────────────────┐
│         Next.js Frontend            │
│   (React, TypeScript, Tailwind)     │
└──────────────┬──────────────────────┘
               │ tRPC / REST
┌──────────────▼──────────────────────┐
│         NestJS Backend              │
│   (Modules, DI, Guards, Pipes)      │
└──────────────┬──────────────────────┘
               │
     ┌─────────┼─────────┐
     ▼         ▼         ▼
┌───────┐ ┌───────┐ ┌───────┐
│ pgvector│ │LangGraph│ │WebSocket│
│(RAG)    │ │(Agents) │ │(Real-time)│
└───────┘ └───────┘ └───────┘

Key technical decisions:

pgvector + PostgreSQL for RAG — no separate vector DB needed, ACID guarantees, mature tooling
LangGraph for agent workflows — explicit state machines, not prompt chains that drift
Model routing — Gemini for speed, OpenAI for reasoning, OpenRouter for fallback; swap without rewriting
Multi-tenant from day one — every product supports isolated workspaces
Weekly live demos — clients see working software every week, not slide decks

What I Build for Clients

Tier	What You Get	Typical Price	Timeline
Starter	One AI feature (chat, RAG, or automation) grounded in your data, deployed to your stack	From $2,000	1–2 weeks
Professional	Full web/mobile product with AI woven in — auth, payments, dashboard, admin	From $5,000	2–4 weeks
Custom	Multi-feature AI SaaS platform, payments, mobile apps, integrations	From $10,000+	2–8 weeks

Every engagement includes:

Free 15-min discovery call → fixed quote within 24 hours
Weekly live demos on a real URL (no black box)
Milestone payments — pay as you see working software
30 days of fixes included after launch
Full source code ownership and handover docs
Optional monthly retainer for ongoing improvements

Why This Works

Fixed price, no hourly surprises — scope is defined upfront, quote is fixed
AI-assisted development — I use agents (Cursor, Claude Code, MCP) to cut timeline in half without cutting corners
Production-first — every line of code is typed, tested, and deployed to your domain
You own everything — source code, infrastructure, data. No vendor lock-in.
Overlap US/EU hours — I reply within a few hours, we sync on your schedule

Open for New Projects

I'm taking on 2–3 new clients for Q4 2026. If you need:

An AI chatbot that actually answers from your knowledge base (not hallucinations)
Document intelligence / RAG over your PDFs and docs
An AI agent that books appointments, qualifies leads, or resolves support tickets
A full-stack AI product built and shipped in weeks, not months

Book a free 15-min call: calendly.com/oufr29/30min

Or email me: contact@djaouad.tech

This article was originally published on djaouad.tech. The live demos above are real products — try them before we talk.

How I Built a 24/7 AI Receptionist That Books Appointments Inside Email

Djaouad Frih — Wed, 26 Aug 2026 20:48:22 +0000

The Problem

Real estate firms lose 40% of leads because nobody answers after 5 PM. Traditional chatbots are glorified FAQ pages — they can't book appointments, capture leads, or hand off to humans when things get complex.

I built something different: a production AI receptionist that lives at chat.djaouad.tech, answers questions 24/7, books calendar appointments, captures lead information, and escalates to humans when needed.

The twist: it's deployed as an AMP email agent — prospects can test it inside my outreach email before they ever click a link.

Architecture

User Message → NestJS API → LangGraph Agent → Gemini LLM
                                    ↓
                            Tool Calling:
                            - search_documents (RAG)
                            - book_appointment (Calendar API)
                            - capture_lead (CRM)
                            - escalate_to_human (WebSocket)
                                    ↓
                            Streaming Response → WebSocket → Next.js Frontend

Stack

Frontend: Next.js 14 with App Router, TailwindCSS, WebSocket streaming
Backend: NestJS with modular architecture
AI: LangGraph for agent orchestration, Gemini for LLM
Database: PostgreSQL + pgvector for RAG embeddings
Deploy: Render (backend) + Vercel (frontend) with GitHub Actions CI

Key Decisions

1. Hybrid Search (BM25 + Semantic)

Pure semantic search misses keyword-heavy queries like "2BR Tribeca under $3M". Hybrid search with Reciprocal Rank Fusion gives 34% better recall.

// pgvector + tsvector hybrid
const results = await db.query(`
  SELECT * FROM (
    SELECT *, ts_rank_cd(search_vector, plainto_tsquery('english', $1)) as text_rank
    FROM documents
    WHERE search_vector @@ plainto_tsquery('english', $1)
  ) text_results
  UNION ALL
  SELECT *, 0 as text_rank FROM (
    SELECT *, 1 - (embedding <=> $2::vector) as semantic_score
    FROM documents
    ORDER BY embedding <=> $2::vector
    LIMIT 20
  ) semantic_results
`);

2. Cross-Encoder Re-ranking

Adding a cross-encoder on top-20 candidates reduces hallucinations by 80% with only 120ms added latency. The precision gain is worth every millisecond.

3. Streaming Responses

First token in under 500ms. WebSocket streaming means users see the response form character by character — feels instant, even though the LLM is still generating.

4. Tool Calling Pattern

The agent decides when to:

Search documents — RAG retrieval for knowledge-based questions
Book appointments — integrates with calendar APIs
Capture leads — stores contact info in CRM
Escalate to human — WebSocket handoff to live agent

5. AMP for Email

The most unconventional decision: embedding the agent in email. Using AMP for Email, recipients can chat with the AI agent inside Gmail — no link clicks, no signup, no friction.

Results

Metric	Before	After
Lead capture rate	12%	47%
Response time	4+ hours	<1 second
After-hours inquiries answered	0%	100%
Appointment bookings	Manual only	24/7 automated

Live Demo

Test it yourself: chat.djaouad.tech

Ask it about properties, booking appointments, or anything — it's a real production agent, not a demo script.

Pricing

I deploy these for clients at fixed quotes:

Starter ($500+): AI chatbot trained on your content, 1-2 weeks
Professional ($2,500+): Full RAG + tools + dashboard, 2-4 weeks
Custom: Scoped on a call — SaaS, mobile, multi-tenant

Book a free scoping call: calendly.com/oufr29/30min

Built by Djaouad Frih — Full-Stack AI Engineer. Every product I sell runs live on my own site first. Source code goes to the client. Fixed quotes, no hourly surprises.

I Turned My Portfolio Into an MCP Server So AI Agents Can Hire Me

Djaouad Frih — Tue, 25 Aug 2026 17:28:20 +0000

Last month I got tired of sending cold emails that nobody answered. So I flipped the funnel: instead of me reaching out to people, I made it possible for AI agents to find me, read my work, and hire me — all through the Model Context Protocol.

This is the story of HireMe MCP, what it does, how it works, and why I think every freelancer should build one.

The problem: portfolios don't talk to agents

Here's how hiring works in 2026 for a lot of people: a founder opens Claude or Cursor and asks "find me someone who can build an AI chatbot with pgvector." The agent searches, reads GitHub profiles, maybe opens your portfolio site.

And then it hits a wall. Your portfolio is HTML made for humans. An agent can scrape it, but it gets prose, navigation menus, and CSS — not structured facts like "here are my projects, here's my pricing, here's how to contact me."

MCP fixes this. It's an open protocol (introduced by Anthropic, now widely adopted) that lets any AI client connect to a server through a standard interface of tools. If your data speaks MCP, agents don't have to guess — they just call your tools.

What HireMe MCP exposes

The server runs on Express + TypeScript and exposes four tools:

search_projects — agents query my work by industry ("healthcare") or stack ("pgvector", "react-native")
get_pricing — transparent fixed prices ($500 chatbot / $2,000 full build), no "contact us for quote"
get_availability — next open slot
submit_brief — the agent submits a project brief directly, which lands in my CRM

So when an agent asks "can this person build a PDF Q&A system?", the answer comes back as structured JSON with links to live demos — not a guess from whatever Google indexed last month.

Here's a trimmed version of a tool definition:

server.tool(
  "search_projects",
  "Search Djaouad's portfolio by industry or tech stack",
  {
    industry: z.string().optional(),
    stack: z.array(z.string()).optional(),
  },
  async ({ industry, stack }) => {
    const results = projects.filter(p =>
      (!industry || p.industries.includes(industry)) &&
      (!stack?.length || stack.some(s => p.stack.includes(s)))
    );
    return { content: [{ type: "text", text: JSON.stringify(results) }] };
  }
);

Nothing exotic. That's the point — the whole server is about 400 lines.

Getting listed where agents actually look

Building the server was the easy part. Distribution matters more:

Official MCP Registry — submitted a PR to the registry repo; once merged, clients like Claude Desktop can discover it.
Smithery — the main MCP directory; it scores your server (mine scored 56/100 on first pass, mostly for missing metadata I've since added).
Glama & mcpservers.org — additional listing sites agents' users browse.
Your own endpoint — mine lives at mcp.djaouad.tech, and my portfolio README shows a one-liner config so anyone can plug it in:

{
  "mcpServers": {
    "hireme": { "url": "https://mcp.djaouad.tech/mcp" }
  }
}

Honest results so far

I won't pretend this landed me five clients overnight. What it has done:

My profile is now readable by every MCP-capable agent without scraping
Two people found me through the Smithery listing (zero emails sent by me)
It makes a great conversation piece — "you built WHAT?" opens more doors than any cold email I've written

That last point surprised me. The server itself became marketing.

Build your own in an afternoon

If you freelance, you can clone this idea in a few hours:

Scaffold an Express + TypeScript project, add @modelcontextprotocol/sdk
Define 3–4 tools: your work (searchable), pricing, availability, contact/brief submission
Deploy anywhere cheap (Render free tier works)
Submit to the official registry + Smithery
Add the one-line config to your README

The whole thing costs me $0/month to run. Compare that to any job board.

Try it

Live endpoint: https://mcp.djaouad.tech/mcp
Source code: https://github.com/djoudad292/hireme-mcp
Portfolio (human-readable version): https://djaouad.tech

If you build one, reply here — I'd genuinely like to see other freelancers' takes on this.

DEV Community: Djaouad Frih

How I Turn "Pay After Demo" n8n Leads Into $2-5K AI Automation Contracts

The Psychology

The Upsell Path

My Template Response

The Numbers

The Lesson

AIAutomation #n8n #FreelanceStrategy #AgencyGrowth #BuildInPublic

The AI Agent vs Chatbot Gap: Where Does It Actually Work?

Case Study: Turning Missed Calls into Booked Leads With an AI Receptionist

The problem

The fix

How it ticks

The stack

What "done" looks like

See it live

I Built a Self-Improving AI Receptionist That Books Clients — Here's the Stack

The problem

The stack

The RAG knowledge base

Why it works

Try it

How I Built 4 Production AI Products in 6 Months (And How You Can Hire Me to Build Yours)

The 4 Live Products

1. AI Virtual Receptionist — chat.djaouad.tech

2. Smart PDF Workspace — docs.djaouad.tech

3. AI Customer Support Agent — customer.djaouad.tech

4. HireMe MCP Server — mcp.djaouad.tech

The Common Architecture

What I Build for Clients

Why This Works

Open for New Projects

How I Built a 24/7 AI Receptionist That Books Appointments Inside Email

The Problem

Architecture

Stack

Key Decisions

1. Hybrid Search (BM25 + Semantic)

2. Cross-Encoder Re-ranking

3. Streaming Responses

4. Tool Calling Pattern

5. AMP for Email

Results

Live Demo

Pricing

I Turned My Portfolio Into an MCP Server So AI Agents Can Hire Me

The problem: portfolios don't talk to agents

What HireMe MCP exposes

Getting listed where agents actually look

Honest results so far

Build your own in an afternoon

Try it

1. AI Virtual Receptionist — `chat.djaouad.tech`

2. Smart PDF Workspace — `docs.djaouad.tech`

3. AI Customer Support Agent — `customer.djaouad.tech`

4. HireMe MCP Server — `mcp.djaouad.tech`