Muhammad.
Kanvis
Muhammad Umar Malik

Muhammad Umar Malik

CS Researcher & Marketing Strategist | Bridging AI, HCI, and Brand Storytelling
Hire Me

Overview

A multidisciplinary engineer and researcher from LUMS with a unique blend of technical depth in AI/HCI and creative leadership in marketing. From building multi-agent fact-checking systems and federated learning models to scaling digital reach by 50% for incubated startups, I specialize in solving complex problems where technology meets human behavior.
Lahore, Pakistan

Experience

Sep 2025 – Apr 2026

WORK

Marketing Head

Cirkles
Lahore, Pakistan
Built the brand’s social media presence from scratch, developing content strategy and overseeing end-to-end production including videography and video editing. Partnered with event organizers and external organizations across Lahore to co-create promotional content for city events. Developed brand guidelines and a content playbook establishing Cirkles’ visual identity across all touchpoints.
•Built the brand’s social media presence from scratch, developing content strategy and overseeing end-to-end production including videography and video editing.
•Partnered with event organizers and external organizations across Lahore to co-create promotional content for city events.
•Developed brand guidelines and a content playbook establishing Cirkles’ visual identity across all touchpoints.

Key Achievements

→Produced a viral documentary-style vertical video campaign featuring artists, achieving a peak of 48.1K views.
→Executed end-to-end marketing and content strategy for 6 major city events in Lahore.
→Grew social presence to 361 followers and 31 posts from a zero-base start.
→Achieved 48.1K views on the top-performing campaign video, demonstrating high organic reach potential.
VideographyVideo EditingContent StrategyBrand Guidelines

Feb 2025 – Dec 2025

WORK

Content Curation Lead

U.n.I
Lahore, Pakistan
Led content strategy and production for a 10,000-user app incubated at the National Incubation Center (NIC), driving a 50.9% increase in accounts reached on Instagram. Handled end-to-end content creation including videography, video editing, and copywriting, producing “build in public” and feature-showcase content across the product lifecycle. Directed a team of interns as head of the marketing function — assigning work, setting creative direction, and managing content production workflows.
•Led content strategy and production for a 10,000-user app incubated at the National Incubation Center (NIC), driving a 50.9% increase in accounts reached on Instagram.
•Handled end-to-end content creation including videography, video editing, and copywriting, producing “build in public” and feature-showcase content across the product lifecycle.
•Directed a team of interns as head of the marketing function — assigning work, setting creative direction, and managing content production workflows.

Key Achievements

→Driving a 50.9% increase in accounts reached on Instagram.
InstagramVideographyVideo EditingCopywriting

Jul 2024 – Oct 2024

WORK

Project Associate (Intern)

ANC Global
Lahore, Pakistan
Acted as the primary bridge between internal stakeholders and developers for a comprehensive website redesign project.
•Owned requirements research and UX scoping, defining the functional needs for the new consultancy platform.
•Coordinated end-to-end communication between ANC leadership and the technical delivery team.

Key Achievements

→Defined functional requirements and scoped UX for 5+ core platform pages, including the Home, Services, and Consultant Directory modules.
→Translated complex business requirements into technical briefs, providing a clear execution roadmap for the external development team.
Stakeholder ManagementTechnical DocumentationUX ScopingRequirements Research

Project Portfolio

ACADEMIC

Present

Automation Bias under Time Pressure

Researcher

Ran a between-subjects randomized experiment (N=129; 645 trials) testing whether an explicit AI confidence bar reduces automation bias under time pressure. Found a 7.6 pp reduction in commission errors (48.7% → 41.1%) in the intervention condition, concentrated in low-confidence AI-wrong trials. Results suggest selective trust calibration rather than blanket distrust — users discount uncertain advice while remaining receptive when AI is reliable.
LLMsUI Intervention
ACADEMIC

Present

Personalized Federated Learning for Cross-Market Portfolio Allocation

Developer/Researcher

Implemented FedAvg and FedProx in PyTorch + Flower across four structurally heterogeneous equity markets (S&P 500, FTSE 100, KSE-100, Nikkei 225) using LSTM portfolios trained directly on Sharpe ratio loss; evaluated over 10 seeds with block bootstrap CIs and Jobson-Korkie significance tests. Found that federation benefit correlates with Hellinger Distance from the global distribution — KSE-100 recovered from a local Sharpe of −0.015 to +0.742 under federation, with 4× variance reduction, confirming FedAvg is non-trivially asymmetric across markets.
PyTorchFlowerLSTMFedAvgFedProx
ACADEMIC

Present

LUMS Campus Gate Access System

Co-developer

Co-developed a role-based Android security platform for campus gate management, delivering 65 user stories across 3 sprints in a team of 5 using agile methodology. Implemented shift handover enforcement blocking closure on unresolved custody, overstay, and approval issues; wrote JUnit repository and ViewModel tests covering boundary conditions.
AndroidJavaFirebaseCameraXZXingMVVMFirestoreJUnit
ACADEMIC

Present

Retrieval-Augmented Generation (RAG Pipeline Optimization)

Developer

Built a RAG pipeline with a systematic hyperparameter sweep across chunk size, overlap, top-K, and similarity threshold on answerable vs. unanswerable question sets. Extended with a prompt caching layer, reducing average response latency by 5.4% while preserving full response consistency.
PythonMistralFAISSLangChain
PERSONAL

Present

Multi-Agent Fact-Checking System

Developer

Built a LangGraph-based adversarial multi-agent pipeline with a researcher, devil’s advocate, neutral verifier, and report writer in a stateful cyclical loop. Benchmarked zero-shot LLM, single RAG agent, and multi-agent crew on a real-world claims dataset; zero-shot outperformed agentic methods on well-known factual claims.
Agentic AIPythonLangGraphLangChainTavily Search
PERSONAL

Present

AI After Dismissal: Generative AI as a Response to Medical Gaslighting

Researcher

Qualitative HCI study investigating how medical dismissal shapes trust in generative AI. Developed 'compensatory trust'—a theoretical construct explaining AI adoption as a response to institutional betrayal. Identified three structural drivers: expert institution failure, high-stakes/inescapable need, and AI accessibility. Key findings: trust is driven by non-dismissal rather than accuracy, and users exhibit 'reliance-without-trust' as a rational hedge when human systems fail. 20+ semi-structured interviews completed.
Generative AIQualitative HCI

Education

May 2027

Bachelor of Science

Lahore University of Management Sciences (LUMS)Specialization in Computer Science

Relevant Coursework

AlgorithmsDeep LearningLarge Language ModelsAI & Machine LearningSoftware EngineeringAI-Driven Software EngineeringOperating SystemsNetwork SecurityBlockchain

Societies & Activities

Society for Promotion and Development of Engineering and Science (SPADES)

Aug 2023

A-Levels

International School LahoreSpecialization in Mathematics, Physics, Chemistry, Business

Impact & Recognition

Awards

MSR 2026 · 2026

Distinguished Mining Challenge Paper Award

Awarded for the paper 'Novice Developers Produce Larger Review Overhead while Vibe Coding'

Society for Promotion and Development of Engineering and Science (SPADES)

Presidential Award

The society’s highest individual honour, given to one member per year.

Patents & Publications

2026 · PUBLICATION

Novice Developers Produce Larger Review Overhead while Vibe Coding

MSR 2026, Rio de Janeiro. Studied 22,953 pull requests from 1,719 vibe coders on GitHub to test whether developer experience still matters in AI-assisted development. Found that lower-experience vibe coders submit 2.15× more commits and receive 4.52× more review comments than experienced ones. Reported 31% lower PR acceptance rates for novice vibe coders.

Governance & Advisory

Society for Promotion and Development of Engineering and Science (SPADES)May 2026 – Present

Executive Officer

COMMITTEE CHAIR

Oversee all day-to-day operations at SPADES as part of the Extended Council, managing the society's core functions and strategic execution.

Overseeing all day-to-day operations of the society.

Serving as a member of the Extended Council to drive strategic execution.

Key Impact
→

Updated Executive Officer role with Extended Council responsibilities and operational oversight.

Society for Promotion and Development of Engineering and Science (SPADES)Jun 2025 – May 2026

Director – Media and Promotions

BOARD MEMBER

Directed the media and promotional strategy for the society's flagship engineering and science events, reaching over 1,600 participants.

Managed 4 distinct brand identities and Instagram accounts: SPADES Main, LSS, PSIFI, and Startup Weekend.

Oversaw outreach and promotional campaigns for a flagship event with 1,600+ delegates.

Key Impact
→

Updated Director role to reflect promotion.

Techstars Startup Weekend LahoreNov 2024 – Mar 2025

Media and Design Coordinator

OTHER

Led visuals and media promotions for Pakistan’s leading entrepreneurial marathon, run with Google for Startups.

FAQs

Frequently Asked Questions

Skills & Interests

Python

Git

Java

PyTorch

NumPy

Pandas

Scikit-Learn

LangChain

LangGraph

C++

Android Studio

Firebase

Adobe Lightroom

DaVinci Resolve

Hugging Face Transformers

Canva

Microsoft Excel

LaTeX

FAISS

C

FFmpeg

Let's Connect

I'd love to hear from you. Feel free to reach out.

These unlock once you finish onboarding and publish your profile.

More people on Kanvis

Browse more people