MUNEEB UR REHMAN QURESHI

AI Engineer | MLOps & Scalable Systems | Ex-DevSnap

Based in: Islamabad, PakistanMost recently: AI Engineer, DevSnapExperience: 1 year

AI Engineer with a track record of deploying production-grade conversational commerce and payment systems. From engineering Stripe Connect architectures at DevSnap to building high-precision MLOps pipelines for industrial anomaly detection, I specialize in bridging the gap between complex ML models and scalable, real-world infrastructure. My experience spans RAG systems, time-series forecasting, and end-to-end model lifecycle management with a focus on measurable business impact.

Experience

AI Engineer

DevSnapworkSep 2025 – May 2026

Summary

Engineered production AI platforms at scale, VOAS AI, a multi-channel conversational commerce (voice, WhatsApp, web, kiosk) serving 5+ concurrent users, processing 1K+ orders and a Car-Detailing Marketplace, two-sided platform connecting clients/detailers with role-aware workflows and real-time order tracking. Full-stack development: React/Next.js, FastAPI, PostgreSQL, Supabase, Docker. Designed and deployed production payment systems: Stripe Connect escrow architecture (15/85 split, 48-hour auto-release, automated payouts); server-side payment processing via Supabase Edge Functions with Stripe webhook integration; self-service detailer onboarding via Stripe Connect Express. Managed payment state, refunds, disputes, and edge cases in production. Built real-time analytics driving business impact: engineered data pipeline tracking 1K+ production orders; identified peak demand patterns and channel preferences through cohort analysis. Deployed production infrastructure: Prometheus/Grafana monitoring, model inference optimization, full ML/system lifecycle ownership (prototype → production → monitoring); CI/CD pipelines, containerized deployment, incident response, cross-functional collaboration.

What I did

Implemented Redis for high-performance caching and session management within the VOAS-AI conversational platform and utilized Firebase for the backend infrastructure of the Car-Detailing Marketplace.

Results

Processed 1K+ orders across multi-channel conversational commerce platforms.
Managed 5-7 concurrent orders during peak rush hours across multiple workspaces on the VOAS-AI platform.
Implemented Stripe Connect escrow architecture with 15/85 split and 48-hour auto-release.
Engineered data pipeline tracking 1K+ production orders for cohort analysis.

ReactNext.jsFastAPIPostgreSQLSupabaseDockerStripe ConnectSupabase Edge FunctionsStripe WebhooksPrometheusGrafanaCI/CDRedisFirebase

Skills

Technical

Python

Docker

Supabase

Time-Series Forecasting

Anomaly Detection

FastAPI

Feature Engineering

Hyperparameter Tuning

Model Evaluation

Cross-Validation

GitHub Actions

CI/CD

Model Versioning

Experiment Tracking

MLflow

Prometheus

Grafana

PostgreSQL

AWS (ECR/ECS)

MySQL

Statistical Testing

Next.js

Ensemble Methods

React

Linux

Redis

Firebase

Areas of expertise

Leadership

Cross-functional Collaboration

Projects

SAHARA - AI-Powered Legal Assistance Platform (FYP)

Lead Developeracademic

Developed an AI-powered RAG chatbot providing legal and Islamic guidance to women using LLaMA 3B, featuring intent-based routing and safety layers for sensitive topics. Indexed 2,000+ legal and islamic documents into a FAISS vector store, achieving 92% retrieval relevance on a 200+ query test set. Deployed on FastAPI with <500ms end-to-end latency for real-time responsiveness.

PythonLLaMA 3BFAISSFastAPIReact

MLOps Pipeline & Predictive Maintenance

personal

Built end-to-end ML pipeline for industrial sensor anomaly detection on 220K samples across 52 channels; Isolation Forest achieved ROC-AUC 0.936 with 18ms inference latency. Achieved 10x reduction in false positives vs. baseline (0.05 FPR); implemented automated drift detection preventing 3 model degradation incidents in production. Engineered three-stage CI/CD pipeline (Docker, AWS ECR/ECS) with MLflow versioning for model lifecycle management. Reduced inference cost by 40% through model optimization; deployed with real-time observability using Prometheus and Grafana.

PythonIsolation ForestMLflowDockerFastAPIGitHub ActionsAWS (ECR/ECS)PrometheusGrafana

Electricity Demand Forecasting

personal

Developed a robust forecasting system for next-day hourly electricity demand across 10 cities, using K-Means clustering to identify customer consumption patterns and LSTM architecture to achieve 4.2% MAPE and 0.85 RMSE. Built an end-to-end engineering pipeline to inform utility resource planning.

PythonK-MeansRandom ForestXGBoostLSTMPandasScikit-learnMatplotlib

Education

BS Data Science

FAST-NUCES, IslamabadData Science2022 – Jun 2026

Coursework

Data WarehousingData MiningMachine LearningBig DataDatabase SystemsStatisticsSQLETLBusiness Intelligence

Societies & activities

Islamabad.AINaSCon

Recognition

Awards

Star of NaSCon '25

FAST-NUCES2025

Recognized as the 'Star' of the National Solution Convention, selected for outstanding contribution and leadership from a pool of 800+ team members.

Patents & publications

On the Impact of Logit Clamping in Knowledge-Enhanced Neural Networks for Citation Classification

publication10.1109/ICIT68548.2026.115776822026

KENN architecture on the CiteSeer citation network; conducted experimental analysis across multiple training data scenarios (10%-90%). Identified a critical implementation flaw in the logit-clamping mechanism, expanding value range to include negative predictions also. Demonstrated a 45% accuracy improvement through methodological refinement.

With M. Hadi, Malaika A.

Read it

Social impact & activities

Lead

Islamabad.AIorganizerPresent

Leading the Isb-chapter of Karachi-AI to foster an AI community through knowledge-sharing, workshops, and expert-led seminars.

Lead Organizer, TechFest '25

TechFest '25workshop2025

Executed an event at Air Uni, with 10+ industry experts and tech workshops with networking sessions for 500+ attendees.

Hire MUNEEBReach out about a role, a contract or a conversation.For recruiters

MUNEEB's twin is AI, it can make mistakes.

More people on Kanvis

Browse more people Create your own Kanvis