Muhammad Mamoon

Muhammad Mamoon

Data Engineer & ML Specialist | HIPAA-Compliant Pipelines | RAG Architectures

Based in: Lahore, PakistanCurrently: Associate Data Engineer, Mountainise Inc.Experience: 2 years

Hire me

Associate Data Engineer specializing in secure healthcare data pipelines and advanced RAG systems. Proven track record in building end-to-end HIPAA-compliant integrations and kernel-space security prototypes. Passionate about scaling AI infrastructure and backend systems.

Experience

Associate Data Engineer

Mountainise Inc.Lahore, PakistanworkMar 2026 – Present

Summary

Leading Cloud Ops, DevSecOps, and Data Engineering for Lucrative.ai, an AI-native CRM suite, while developing full-stack features and AI pipelines.

What I did

  • Architecting and maintaining data pipelines for Lucrative.ai, ingesting data from HubSpot, Salesforce, and Airtable into ClickHouse.
  • Managing the transition from AWS serverless infrastructure to Azure Kubernetes Service (AKS), including cluster setup, ACR settings, and secret management.
  • Developing full-stack features using Next.js and Python, and managing containerized deployments across testing, staging, and production environments.
  • Rewriting core libraries and middleware to eliminate technical debt and ensure the application remains infrastructure-agnostic during cloud migrations.
  • Implementing DevSecOps practices and rate-limited ingestion logic to ensure data reliability and security.
  • Designed and implemented an automated HIPAA-compliant workflow integrating JotForm, Microsoft Fabric, Power Automate, and SharePoint for patient document delivery.

Results

  • Engineered a multi-source data ingestion pipeline for Lucrative.ai, migrating from DuckDB to ClickHouse to resolve a critical single-concurrency bottleneck and enable high-speed, multi-concurrent database calls.
  • Orchestrated a major infrastructure migration from AWS serverless to Azure Kubernetes Service (AKS), refactoring core middleware and library handlers to decouple the application from provider-specific services.
  • Architected a secure Model Context Protocol (MCP) tunnel for Lucrative.ai, implementing a 'Human-in-the-Loop' verification system that requires explicit user approval and re-authentication for all state-changing AI actions.
  • Automated a manual healthcare document generation and delivery process, replacing manual transcription from JotForm to Word templates with an end-to-end pipeline using Microsoft Fabric, Power Automate, and SharePoint.
  • Engineered a HIPAA-compliant data cleaning and mapping layer in Microsoft Fabric to ensure data integrity before document generation.
  • Reduced document processing time from ~2 hours to under 5 minutes per patient record, achieving a 95%+ efficiency gain.
  • Automated end-to-end HIPAA-compliant delivery for a Kansas-based healthcare provider, eliminating manual transcription errors.
PythonSQLn8nSnowflakeAWSAzureMicrosoft FabricJotformAWS BedrockHIPAAClickHouseDuckDBKubernetesAzure Kubernetes Service (AKS)Next.jsDevSecOpsCloud OpsHubSpot APISalesforce APIAirtable APIModel Context Protocol (MCP)Human-in-the-Loop (HITL)AI SecurityAntigravityClaude Code

Full Stack Engineer

CirklesworkAug 2025 – May 2026

Summary

Developed a full-stack platform using Next.js App Router and TypeScript; implemented Zod schema validation to enforce API contracts. Implemented server-side data fetching and caching, with search and pagination workflows.

What I did

  • Optimizing Django backend performance by identifying and resolving N+1 query issues.
  • Designing and building a complex, interactive form builder with drag-and-drop functionality and conditional logic.
  • Revamping the frontend architecture using Next.js and React, focusing on pixel-perfect UI and performance optimizations.

Results

  • Engineered a custom drag-and-drop form builder (similar to Tally) supporting conditional logic, reordering, and 10+ field types for event registration.
  • Led a comprehensive frontend and backend revamp, reducing page load times from 7 seconds to under 2 seconds by optimizing Django N+1 queries and implementing Next.js best practices.
  • Implemented strict frontend schema validation using Zod to eliminate junk data and reject invalid submissions before they reach the backend, ensuring high data integrity.
  • Leveraged agentic AI tools (Antigravity, Claude Code) to accelerate the Cirkles revamp while maintaining pixel-perfect UI standards.
  • Reduced page load times by ~70% (from 6-7s to 1-2s) across the entire platform.
Next.jsTypeScriptNode.jsTailwind CSSZodDjangoPerformance OptimizationAPI DesignAntigravityClaude CodeAgentic AI Workflow

ML Engineer

U.n.I / Kanvis.AIworkSep 2024 – Dec 2025

Summary

Designed a three-stage retrieval-augmented generation pipeline and integrated named-entity recognition to extract entities and interpret ambiguous queries. Developed backend APIs for posts, comments, connections, and token-based access.

What I did

  • Utilized spaCy and embedding-based chunking to build a high-precision search engine for professional networking.

Results

  • Architected a three-stage RAG pipeline using spaCy NER and classification to automate user matching based on complex natural language queries (e.g., specific graduation years and majors).
  • Implemented a keyword-extraction and embedding-search workflow that processed intent to return the top 10-15 most relevant professional connections.
  • Achieved ~80-90% retrieval accuracy for networking queries after tuning the three-stage RAG pipeline and NER classification.
Hugging FaceDjango REST FrameworkPostgreSQLRAGNERspaCyNLPEmbeddings

Skills

Technical

GitGit
PythonPython
LLM APIs
REST APIs
RAG
PostgreSQLPostgreSQL
LinuxLinux
JavaScriptJavaScript
SQLSQL
TypeScriptTypeScript
NLP
NER
Hugging Face
n8n
Django REST FrameworkDjango REST Framework
FastAPIFastAPI
Next.jsNext.js
Snowflake
Azure
C/C++C/C++
AWS
Kernel Modules
Node.jsNode.js
Java
PySide6
GCPGCP
Power BI
Sigma
ClickHouse
DuckDB
KubernetesKubernetes
Azure Kubernetes Service (AKS)Azure Kubernetes Service (AKS)
ElevenLabs
Data Augmentation
Random Forest
DevSecOps

Languages

English

Projects

Qanun AI | Legal Question Answering

Project Leadpersonal
Led development of a legal AI assistant for Pakistani law using a multi-stage retrieval pipeline over more than 300 legal documents. Implemented document chunking, semantic retrieval, query refinement, and response-validation stages to ground answers in retrieved material.
Gemini APIVector DatabasesFastAPI

JRXML-Aware Document Export Engine

Developerpersonal
Developed a wrapper that interprets JRXML report structure for Word export, addressing layout handling for bands, text fields, tables, and line breaks through client feedback.
JavaJasperReportsJRXML

SafeKeys | Kernel-Space Keylogger Detection

Undergraduate Researcheracademic
Built an end-to-end Linux prototype for kernel-space keylogger detection using an ensemble of XGBoost, LightGBM, and Random Forest. Developed a detection logic based on behavioral metadata—monitoring process access frequency to the input subsystem—to identify kernel-level threats that disguise themselves as legitimate processes. Conducted extensive research across malware forums and archives to build a robust training dataset.
LinuxPythonPySide6XGBoostLightGBMRandom ForestKernel Instrumentation

AI Nexus Hackathon (LUMS)

personal
Trained a machine learning model to classify 25 different Harry Potter spells based on hand-based movement data during a four-day hackathon.
Random ForestCSV

Urdu Voice-Based Farming Q&A Agent

personal
Built a voice-based Q&A agent for a local Pakistani farming app using ElevenLabs' voice agents and a vector database of agriculture datasets to allow farmers to converse naturally in Urdu.
ElevenLabsVector Database

Education

BS (Hons.) Computer Science

Forman Christian College (A Chartered University)Computer ScienceGPA 3.6792022 – 2026

Thesis. SafeKeys: Kernel-Space Keylogger Detection

Societies & activities

Vice Rector's List: Spring 2023, Spring 2025, Fall 2025

Recognition

Awards

Winner, AI Nexus Hackathon

LUMS

National-level competition · PKR 75,000 prize. Built a spell-recognition system using LSTM and Naive Bayes classifiers; developed a data-augmentation pipeline expanding 5,000 source samples to 500,000 examples.

Certifications

IELTS Academic: 8.0 overall

IELTS2026

What drives the work

Motivations

  • High-agency generalist architecture
  • Research-oriented problem solving
  • Continuous learning across broad technical domains
Hire MuhammadReach out about a role, a contract or a conversation.For recruiters

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

More people on Kanvis

Browse more peopleCreate your own Kanvis