AI Backend Engineer · Karachi, Pakistan

Muhammad
Waris

I build the intelligence layer
behind your product.

Production-grade AI backends, conversational assistants, RAG-powered knowledge systems, and intelligent automation pipelines — engineered to scale and built to last.

I also build AI chatbots, AI assistants, abandoned-cart recovery, and WhatsApp automation for businesses across industries — stores, restaurants, travel agencies, salons, clinics, coaching centers, and more.

See Flintel → All Projects Download Resume ↓

Scroll to explore

engineer.profile.json

6⁺

Production
Projects

8⁺

Certifications
Earned

8⁺

Stack
Layers

2^yr

Building
AI Systems

Technical Expertise

What I Build With

I'm an AI backend engineer based in Karachi, Pakistan, building production-grade conversational AI, RAG pipelines, and automation backends for businesses across fintech, e-commerce, and hospitality. Alongside client work, I'm building Flintel — a live signal-intelligence product for fintech sales teams — and completing an AI engineering internship at FlyRank AI.

01 — Backend

Backend Engineering

FastAPI REST APIs Microservices JWT Auth Webhooks

02 — AI/ML

AI & Machine Learning

OpenAI API RAG Systems LangChain Prompt Eng.

03 — Database

Database & Storage

MongoDB Atlas NoSQL Design Vector Search

04 — Chatbots

Chatbot Platforms

WhatsApp API AI Assistants WebSockets Session Memory

05 — Languages

Languages

Python SQL Bash

06 — DevOps

DevOps & Deployment

Git GitHub Render Cloud CI/CD

Live Product · Founder & Builder

Flintel —
Signal Intelligence
for Fintech & FX

This isn't a case study — it's a real, live product in production today, currently deployed with a Canadian fintech expanding to 25 markets. Flintel monitors Twitter/X, Reddit, Telegram, and competitor reviews around the clock. The moment a buyer posts publicly about a cross-border payment or FX problem your product solves, Flintel scores their intent, identifies who they are, drafts the outreach message, and delivers the full signal to Slack and HubSpot — in under 90 seconds.

Three-layer pipeline — Listener, Filter, Intelligence — with a two-stage filter that clears 95% of noise before AI ever touches it
Intent scoring 0–10 with buyer identification, routed automatically: instant alert, daily digest, silent watchlist, or discarded
AI-generated, ready-to-send outreach script attached to every qualified signal
Slack + HubSpot delivery in under 90 seconds — live in production, not a concept
Built for FX & remittance platforms, crypto exchanges, and B2B payments companies — anywhere buyers vent publicly before they switch

Visit the Live Product → Discuss a Use Case

flintel.signal_feed · live Live

Twitter/X · 23s ago

Founder flagged a blocked cross-border wire payment and asked for a better corridor solution.

Intent Score 9.2 Script Ready ✓

Reddit · 61s ago

Business owner venting about a failed group booking, actively asking for alternatives.

Intent Score 8.7 Sent to Slack ✓

<90s

Signal Delivery

Platforms Monitored

24/7

Always Running

Selected Work

Production Projects

🛒

Featured Live FastAPI + OpenAI

AI E-Commerce Conversational Assistant

A production-grade conversational commerce backend processing natural language through an 8-layer retrieval intelligence pipeline. Multilingual NLP (Urdu + English), real-time WebSocket communication, composite ranking engine, and persistent session memory.

Multilingual NLP with Urdu Unicode detection and synonym mapping
Multi-dimensional ranking: keyword, category, color, material & trending signals
WebSocket-driven real-time interface — zero HTTP handshake overhead per message
Behavioral analytics pipeline for continuous ranking calibration
Deployed live on Render Cloud with MongoDB Atlas persistence

Live App → Info Site How It Works →

📱

Live WhatsApp API

WhatsApp Restaurant Reservation Bot

End-to-end WhatsApp automation — customers reserve tables, place orders, and receive real-time confirmations entirely through chat. Includes full CRM dashboard for restaurant staff.

Automated message flows with cancellation handling
Staff CRM dashboard with JWT authentication
FastAPI + MongoDB Atlas on Render

CRM Dashboard →

🧠

Live RAG + OpenAI

RAG-Based AI Knowledge System

Retrieval-Augmented Generation system grounding OpenAI responses in ingested documents. Custom document pipeline with vector-based retrieval for precise, factual answers.

Custom ingestion pipeline and vector retrieval
OpenAI integration for grounded responses
Deployed as live web application

Live App →

✈️

Live FastAPI + OpenAI

Travelfy — AI Travel Concierge

A conversational AI travel concierge that plans trips end-to-end through natural chat — understanding traveler intent and generating tailored recommendations and itineraries in real time, backed by the same FastAPI + OpenAI + MongoDB architecture as the e-commerce assistant.

Conversational trip planning with contextual, multi-turn follow-ups
Session memory for ongoing traveler preferences
MongoDB Atlas persistence on Render Cloud

Live App →

🎬

API Integration Live Platform

Pixup8k Platform Integration

Backend API integration for the Pixup8k media platform. REST API development handling media workflows and automated processing pipelines with full authentication and webhook support.

REST API with JWT authentication and webhook support
Media workflow automation and processing pipelines

Live Platform →

System Design Depth

8-Layer
Architecture

The AI E-Commerce Assistant isn't a wrapped ChatGPT call — it's a purpose-built retrieval intelligence system. Eight discrete layers handle everything from WebSocket transport to behavioral analytics, working in concert to deliver sub-second, contextually aware responses across Urdu and English.

01 — Transport

WebSocket Layer

Persistent bidirectional connections. Zero re-handshake overhead per message. Real-time streaming throughout sessions.

02 — Runtime

FastAPI Async Layer

ASGI non-blocking I/O. Concurrent session management without threading overhead.

03 — Language

NLP Query Layer

Urdu + English detection. Category, color, material extraction. Intent classification via pattern matching.

04 — Memory

Session Memory

Per-session state with multi-turn context. Preference history enables contextual follow-up generation.

05 — Retrieval

Product Retrieval

NLP signal → MongoDB filter translation. Querying against category, color, material, and price constraints.

06 — Scoring

Ranking Engine

6-dimension composite score: keyword relevance, category, color, material, rating, and trending weight.

07 — Telemetry

Analytics Pipeline

Behavioral event stream: searches, clicks, views, price signals, language patterns. Live feedback loop for calibration.

08 — UX

Dynamic Suggestions

Contextual follow-ups based on session state. 2–4 relevant refinements per turn guiding discovery flow.

Backend Engineering

AI & Machine Learning

Database & Storage

Chatbot Platforms

Languages

DevOps & Deployment

Flintel —Signal Intelligencefor Fintech & FX

AI E-Commerce Conversational Assistant

WhatsApp Restaurant Reservation Bot

RAG-Based AI Knowledge System

Travelfy — AI Travel Concierge

Pixup8k Platform Integration

WebSocket Layer

FastAPI Async Layer

NLP Query Layer

Session Memory

Product Retrieval

Ranking Engine

Analytics Pipeline

Dynamic Suggestions

Chatbot Development Guide 2025 — Beginner + Advanced

AI / Python Backend Development

API Development & FastAPI Training

Claude 101

Introduction to Claude Cowork

AI Fluency: Framework & Foundations

AI Fluency: AI Capabilities & Limitations

AI Fluency for Students

Backend AI Engineering Internship

Muhammad
Waris

Flintel —
Signal Intelligence
for Fintech & FX