OPEN TO NEW CHALLENGES

Engineering production-grade AI systems that turn data, voice, and vision into real business outcomes.

I build end-to-end AI products — from multi-agent LLM pipelines and computer-vision suites to real-time telephony platforms — and ship them to production.

Explore Case Studies Get in Touch

FastAPIGoPyTorchKafkaFreeSWITCHVueNext.jsAWSAzure

Expertise

Tools I ship with

From model training to VoIP routing — whatever the product needs to get to production.

AI & LLM Systems

GPT / Claude / Gemini APIsRAG PipelinesLangChainVector DBs (Pinecone / FAISS)Whisper (ASR)OCR (Tesseract / EasyOCR)Prompt Engineering

Machine Learning & Vision

PyTorch / TensorFlowScikit-learnTime-series ForecastingYOLO / OpenCVFace Recognition & Anti-SpoofingGenerative Image / Video (VEO3, WAN 2.2)

Backend & Data

Python / FastAPIGoPostgreSQL / TimescaleDBRedisApache KafkaWebSocketsJWT / OAuth2

Frontend, Cloud & DevOps

Vue 3 / PrimeVueReact / Next.jsTypeScript / TailwindDocker / GitHub ActionsAWS S3 / AzureFreeSWITCH / Telnyx (VoIP)n8n / Make / Zapier

Projects

Production AI Case Studies

Real deployments of 10 AI systems — each with the problem, the build, and what it changed.

Telnyx SIP → FreeSWITCH → Go router

→ FastAPI → Kafka → TimescaleDB

→ live WebSocket dashboard

2025 – PresentTelephony · Real-time Analytics

Pay-Per-Call Tracking & Routing Platform

The Challenge

Performance-marketing businesses needed to know which ad generated each inbound phone call, route it to the best buyer in milliseconds, record it, and see live analytics — the same problem Ringba solves at enterprise scale.

Technical Solution

Built a full Ringba-class platform: FastAPI backend with PostgreSQL + TimescaleDB, a Go dial-plan router that answers FreeSWITCH mod_xml_curl requests from Redis alone, Telnyx SIP trunking, Kafka event pipelines, a sandboxed QuickJS/Deno sidecar for operator JavaScript, and a Vue 3 + PrimeVue dashboard with live WebSocket call boards.

Strategic Outcome

Delivered campaigns, tracking numbers, ring trees, real-time bidding targets, caps, postbacks, and a tenant-isolated billing service — all covered by property-based, contract (Schemathesis) and Playwright test suites and CI/CD.

FastAPIGoFreeSWITCHTelnyxKafkaTimescaleDBRedisVue 3Docker

Flowchart of the two-phase AI cartoon video generation pipeline

2025Generative AI · Video

AI Cartoon Video Generation Pipeline

The Challenge

A YouTube channel needed a steady stream of consistent cartoon episodes, but producing character-accurate scenes and videos by hand took hours per clip.

Technical Solution

Designed a two-phase pipeline: Phase 1 generates scenario-based character images from a base image (GPT Image 1, later Nano Banana via Replicate); Phase 2 refines the raw prompt with the GPT API and routes to VEO3 (paid) or WAN 2.2 (free) for video synthesis.

Strategic Outcome

Turned a base character image plus a one-line scene description into a finished cartoon video in minutes, with a cost-aware paid/free rendering switch.

ReplicateGPT APINano BananaVEO3WAN 2.2Python

2024LLM · Web Scraping · NLP

Automated Toxicology Report Generator

The Challenge

Safety professionals spent days compiling toxicology dossiers for chemical compounds from scattered scientific and regulatory sources.

Technical Solution

Built an AI system that takes a CAS number or compound name, scrapes PubMed, Google Scholar, NIOSH, FDA and IARC, filters irrelevant content with NLP, and compiles structured findings (in vivo / in vitro results, natural occurrence), regulatory classifications and risk assessments with citations.

Strategic Outcome

Reports and raw source data are stored in AWS S3 for traceability, giving researchers a cited, reproducible report in minutes instead of days.

PythonNLPWeb ScrapingLLMAWS S3

2024Workflow Automation · CRM

Earlibird AI Automation Platform

The Challenge

Small businesses were losing leads to slow follow-ups and manual scheduling, with customer data scattered across channels.

Technical Solution

Integrated Make.com, n8n, Zapier and GoHighLevel to automate lead capture, scheduling, follow-ups and support across SMS, email, Meta (FB/IG) and live chat, with CRM sync, contact reactivation, AI-driven web forms and centralized funnels + reporting.

Strategic Outcome

Faster response times, higher conversions and significantly reduced operational overhead for the business.

n8nMake.comZapierGoHighLevelMeta APIsCRM

AI-powered crypto and forex trading bot overview

2024ML · Time-series · Trading

AI-Powered Crypto & Forex Trading Bot

The Challenge

Traders needed reliable, data-driven predictions across both crypto and forex markets instead of gut-feel entries.

Technical Solution

Trained ML models on large-scale historical data from Binance and OKX (BTC, ETH, XRP, SOL, DOGE, SUI) plus key forex pairs for hourly, daily and weekly trend prediction, with future-trend simulation and automated execution based on confidence thresholds.

Strategic Outcome

A single bot supporting diversified crypto + forex portfolio management with configurable strategy parameters.

PythonScikit-learnPyTorchBinance APIOKX APIPandas

2024NLP · OCR · ASR · RAG

Political Sentiment Chatbot

The Challenge

News and media analysts needed to track political sentiment across TV tickers, audio debates and online articles — three completely different data modalities.

Technical Solution

Built an insight engine that OCRs news tickers (Tesseract / EasyOCR), transcribes audio with OpenAI Whisper, scrapes articles, stores structured data in PostgreSQL and embeddings in a vector DB (Pinecone / FAISS), and runs a fine-tuned transformer for sentiment and topic analysis.

Strategic Outcome

Sentiment classification, trending-topic and named-entity detection, and ticker tracking with summarization of evolving narratives — all queryable through a chat interface.

WhisperTesseractPostgreSQLPineconeTransformersPython

Face Solutions computer vision feature overview

2023Computer Vision · Deep Learning

Face Solutions Suite

The Challenge

A client needed one unified engine for every face-related task — from security verification to privacy protection — across video, image and photo inputs.

Technical Solution

Engineered a modular CV suite: face detection, recognition/verification, age & gender prediction, anti-spoofing, deep-learning face swap, face blur for privacy, and image enhancement, with multi-format input processing.

Strategic Outcome

Eight production-ready modules behind a single API, reusable across security, media and analytics use cases.

OpenCVPyTorchDeep LearningFastAPIPython

AI-powered real estate price and market prediction bot

2023ML · Predictive Analytics

AI Real Estate Price & Market Prediction Bot

The Challenge

Realtors and investment groups priced properties on intuition, with no way to model how a listing compared to competitors or its likelihood of selling.

Technical Solution

Built a prediction engine trained on sale-price trends, neighbourhood demand, property type/size and time-on-market records, factoring in location value, competitor listings and supply-demand balance.

Strategic Outcome

Flags over/under-priced listings, models sale probability within a timeframe, and runs impact analysis of price changes on sales probability.

PythonScikit-learnXGBoostPandasGeo Data

AI-powered autonomous driving and traffic monitoring solution

2023Computer Vision · Edge AI

AI-Powered Autonomous Driving Solution

The Challenge

The National Highway & Motorway Authority needed real-time traffic monitoring and safety enforcement on live highway video.

Technical Solution

Combined edge AI with real-time video processing for lane detection, traffic-sign recognition, vehicle detection & classification, license-plate recognition, speed estimation and environmental context analysis (weather / temperature).

Strategic Outcome

An intelligent, scalable and cost-effective solution for smart highway infrastructure.

YOLOOpenCVEdge AIOCRPython

Crypto Pulse Unlimited data flow diagram

2023NLP · Data Engineering · Cloud

Crypto Pulse — AI Sentiment Pipeline

The Challenge

Crypto sentiment lives across Reddit, Discord, YouTube, news and Binance Square — too noisy and fragmented to read manually.

Technical Solution

Built a five-stage pipeline: multi-source ingestion with scheduled streaming/batch collection, cleaning & deduplication, OCR + translation + normalization, GPT-based sentiment analysis with signal aggregation, and Azure deployment via a Flask REST API + Streamlit frontend.

Strategic Outcome

Live bullish / bearish / neutral predictions with confidence scores, historical tracking, trend analysis and exportable reports.

GPTFlaskStreamlitAzureOCRPython

Open Source

On GitHub

github.com/MuhammedEman14

2026Python · JavaScript

MedicalImagingandRadiology

A Flask-based medical image classification web app that supports both pre-trained and fully customizable models for a range of diseases, with built-in dataset management and model-training workflows.

FlaskDeep LearningMedical ImagingModel Training

View repository

2025JavaScript

Techmart

Full-stack e-commerce analytics platform delivering real-time business intelligence, customer-behaviour analytics, fraud detection and inventory management — with ML for customer segmentation, lifetime-value prediction, churn detection and personalised product recommendations.

Full-StackAnalyticsMLFraud Detection

View repository

2025Jupyter Notebook

NLP Based Recommendation through Citation and Metadata Analysis

Research-paper recommender that tackles the explosion of academic literature: combines NLP over abstracts, citation-graph analysis and metadata to surface the most relevant studies, with visualisations to explore the results.

NLPCitation AnalysisRecommender Systems

View repository

2025Python

Image Reconstruction Using Pixel RNN

Trained a PixelRNN generative model to reconstruct occluded regions of images pixel by pixel; the attached report documents the architecture, training and reconstruction results.

PixelRNNGenerative ModelsComputer Vision

View repository

2024TypeScript · JavaScript

Automatic Test Case Generator

Node.js + TypeScript application with a web frontend that automatically generates test cases for user-provided TypeScript functions, runs them, and extracts code-coverage metrics from the results.

Node.jsTypeScriptTest AutomationCode Coverage

View repository

2024Jupyter Notebook · Python

Eye Disease Detection

Eye-disease classifier built on a VGG16-based TensorFlow/Keras model, served through a Flask API with a web frontend for uploading retinal images and getting instant predictions.

TensorFlowVGG16FlaskMedical AI

View repository

Muhammad Eman

Full Stack AI Engineer

Pakistan

About

Research driven. Production minded.

I'm a Full Stack AI Engineer who likes owning the whole stack: the model, the pipeline that feeds it, the API that serves it, and the UI people actually touch.

My work spans multi-agent LLM systems, retrieval pipelines, computer vision, time-series forecasting, and — most recently — a real-time telephony platform where a bad millisecond means a dropped call.

I care about systems that survive production: typed contracts, property-based tests, observability, and clean boundaries between services.

“If it isn’t tested, observable, and deployed, it isn’t done.”

Experience

SageTeck

2025 – Present

Full Stack AI Engineer

Building a Ringba-class pay-per-call tracking and analytics platform: tracking numbers, campaigns, ring trees, real-time bidding targets, caps and postbacks.
Architected the call path — Telnyx SIP → FreeSWITCH → Go dial-plan router (Redis-only, sub-millisecond) → FastAPI backend — with Kafka consumers writing CDRs to TimescaleDB and pushing live events over WebSockets.
Shipped a sandboxed QuickJS/Deno sidecar for operator JavaScript, a tenant-isolated billing service, and a Vue 3 + PrimeVue dashboard, backed by property-based, Schemathesis contract and Playwright test suites in CI/CD.

Brainbox Automations

2024 – 2025

AI Engineer

Earlibird AI Automation Platform — automated lead capture, scheduling, follow-ups and support across SMS, email, Meta and live chat using n8n, Make.com, Zapier and GoHighLevel with CRM sync and centralized reporting.
Automated Toxicology Report Generator — AI system that scrapes PubMed, NIOSH, FDA and IARC for a compound, filters with NLP, and compiles cited toxicological findings and risk assessments stored in AWS S3.
AI Cartoon Video Generation Pipeline — two-phase generative pipeline (Nano Banana / GPT Image → VEO3 / WAN 2.2) turning a base character and a scene prompt into finished YouTube cartoon videos.