Wasif.
Kanvis
Wasif Ahmed

Wasif Ahmed

Electrical Engineering at LUMS | Edge AI & Model Compression Specialist | Chief Research Officer @ SPADES
Hire Me

Overview

Electrical Engineering student at LUMS with a specialized focus on Edge AI, model compression, and high-stakes engineering leadership. As Chief Research & Initiatives Officer at SPADES, I oversee international-scale STEM projects, including Pakistan’s first student rocketry initiative. Expert in PyTorch, quantization (QAT/PTQ), and pruning, with a proven track record of formulating complex ML problems into deployable, high-performance solutions.
Lahore, Pakistan

Experience

WORK

Chief Research & Initiatives Officer

SPADES: Pakistan’s Largest Student STEM Society
LUMS, Lahore
Lead ambitious engineering projects including an FPV Drone, Rocketry, Raptor Jet, and a Robotic Arm 3D Printer; spearheading Pakistan’s first student rocketry project to compete internationally by 2027, with first test launch planned for March 2025. Strategically drive society’s external relations, partnerships, and collaborations with industry leaders to elevate SPADES’ visibility and outreach. Oversee major society events delivering high-value experiences: LUMS Science Symposium (LSS) — a 5-day research-based event hosting 500+ high school students from across Pakistan; PSIFI — society’s flagship high school olympiad with 1,500+ participants; HackX — university hackathon with 200+ participants. Oversee Techstars Startup Weekend LUMS (TSWL), Cosmos 101 (astronomy course), the Observatory, and other society initiatives. Secured an invitation for PSIFI’s top-performing team to represent Pakistan at the Asian Physics Olympiad (Saudi Arabia), strengthening SPADES’ national academic recognition. Organized a month-long initiative teaching coding and mathematics to 100+ underprivileged students. Collaborated with environmental organizations to distribute smog detectors across 10 schools in Lahore, raising air-pollution awareness. Hosted Pakistan’s first NASA engineer talk, highlighting the James Webb Space Telescope project to 300+ attendees. Partnered with UPLIFT AI (US-based) to organize internships for 33 students; mentored on RAG, APIs, and custom RAG systems to build applied AI projects. Supported UPLIFT AI in collecting native-language data for Pakistan’s first GPT-based WhatsApp voice bot for low-literacy users (voice banking, agriculture guidance, bill payments).
Lead ambitious engineering projects including an FPV Drone, Rocketry, Raptor Jet, and a Robotic Arm 3D Printer
Spearheading Pakistan’s first student rocketry project to compete internationally by 2027, with first test launch planned for March 2025
Integrating gyroscope and accelerometer sensors for real-time flight stabilization and telemetry data logging
Strategically drive society’s external relations, partnerships, and collaborations with industry leaders to elevate SPADES’ visibility and outreach
Oversee major society events delivering high-value experiences: LUMS Science Symposium (LSS) — a 5-day research-based event hosting 500+ high school students from across Pakistan
Oversee PSIFI — society’s flagship high school olympiad with 1,500+ participants
Oversee HackX — university hackathon with 200+ participants
Oversee Techstars Startup Weekend LUMS (TSWL), Cosmos 101 (astronomy course), the Observatory, and other society initiatives
Secured an invitation for PSIFI’s top-performing team to represent Pakistan at the Asian Physics Olympiad (Saudi Arabia), strengthening SPADES’ national academic recognition
Organized a month-long initiative teaching coding and mathematics to 100+ underprivileged students
Collaborated with environmental organizations to distribute smog detectors across 10 schools in Lahore, raising air-pollution awareness
Hosted Pakistan’s first NASA engineer talk, highlighting the James Webb Space Telescope project to 300+ attendees
Partnered with UPLIFT AI (US-based) to organize internships for 33 students
Mentored on RAG, APIs, and custom RAG systems to build applied AI projects
Supported UPLIFT AI in collecting native-language data for Pakistan’s first GPT-based WhatsApp voice bot for low-literacy users (voice banking, agriculture guidance, bill payments)

Key Achievements

Leading Pakistan’s first student rocketry project with a roadmap to international competition by 2027
Targeting March 2025 for the first test launch focused on high-fidelity post-flight data logging and telemetry analysis using IMU sensors (gyroscope, accelerometer)
Developing the Raptor Jet project with a primary engineering focus on maximizing top speed and aerodynamic efficiency
Winner of CERN Beamline Competition (2023) out of 367 teams globally
Secured invitation for PSIFI’s top-performing team to represent Pakistan at the Asian Physics Olympiad
Organized month-long initiative for 100+ underprivileged students
Partnered with UPLIFT AI to organize internships for 33 students
Winner of CERN Beamline Competition (2023) out of 367 teams globally
Secured invitation for PSIFI’s top-performing team to represent Pakistan at the Asian Physics Olympiad
Organized month-long initiative for 100+ underprivileged students
Partnered with UPLIFT AI to organize internships for 33 students
Targeting 100 m/s top speed for the Raptor Jet project through aerodynamic optimization
FPV DroneRocketryRaptor JetRobotic Arm 3D PrinterRAGAPIsGPT-based WhatsApp voice botIMU SensorsGyroscopeAccelerometerTelemetryData Logging

Project Portfolio

ACADEMIC

Present

RL-ClipQAT: Reinforcement Learning for Dynamic Mixed-Precision QAT

Course/Research Project

Formulated mixed-precision quantization-aware training as a finite-horizon MDP; trained an actor–critic policy to assign per-layer bit-widths from {4,6,8} under an average-bit budget. Integrated QAT (STE / fake-quant) with learnable clipping modes; used a frozen FP32 teacher for distillation and KL-divergence sensitivity signals. Best policy reached 5.75 average bits while matching uniform 8-bit accuracy (per project results).
Reinforcement LearningActor-CriticQATSTEFake-quantKnowledge DistillationKL-divergence
ACADEMIC

Present

Weight and Activation Quantization (K-means, Linear)

Student

Performed per-layer K-means weight quantization (8-bit/256 centroids, 4-bit/16 centroids) for dense and pruned models. Implemented centroid QAT: treated centroids as trainable parameters, accumulated gradients from effective quantized weights, and optionally alternated assignment ↔ centroid updates. Built linear quantization pipelines: static per-tensor (weights+activations) vs dynamic activations with per-channel weights; compared accuracy and profiling metrics.
K-meansQuantizationCentroid QATLinear quantizationStatic quantizationDynamic quantization
ACADEMIC

Present

Multi-student Distillation Ensemble

Student

Implemented multi-student distillation. Trained sub-students on representation slices and built an ensemble with frozen backbones and a fine-tuned classifier head.
Multi-student distillationEnsemble learningRepresentation slices
PERSONAL

Present

Pruning VGG16-BN for Edge Deployment

Student

Benchmarked baseline models (size, MACs, latency, memory, energy) and evaluated Top-1/Top-5 accuracy on CIFAR-10/100. Implemented unstructured pruning via layerwise sensitivity sweeps and fine-tuning; generated sparse masks/COO tensors and analyzed accuracy–efficiency trade-offs. Implemented iterative saliency-based pruning (GrasP-style) with staged pruning targets during training. Implemented structured channel pruning via regression/LASSO-based channel selection and squares weight reconstruction; pruned sequentially across convolutional blocks and fine-tuned.
VGG16-BNCIFAR-10CIFAR-100Unstructured pruningCOO tensorsGrasPStructured channel pruningLASSO
PERSONAL

Present

Knowledge Distillation for Compact CNNs (Logit-matching, FitNets)

Student

Implemented KD logit-matching (KL divergence with temperature scaling) to distill VGG16-BN → VGG11-BN on CIFAR-100. Analyzed rank-mismatch limitations of magnitude-based KL matching using toy logits and real CIFAR-100 outputs; tested logit standardization for stability. Implemented hint-based distillation (FitNets) using intermediate feature supervision with functional regressor alignment and two-stage training.
Knowledge DistillationLogit-matchingKL divergenceVGG16-BNVGG11-BNFitNetsFunctional regressor alignment

Education

2027

B.S. Electrical Engineering

Lahore University of Management Sciences (LUMS)Specialization in Electrical Engineering

Relevant Coursework

AI on Edge Devices (AI624)

Societies & Activities

SPADES

Impact & Recognition

Awards

CERN · 2023

CERN Beamline Competition Winner

Winner out of 367 teams globally

National Physics Talent Contest

National Physics Talent Contest (NPTC): Top 50 in Pakistan

Achieved twice

International Astronomy and Astrophysics Competition

International Astronomy and Astrophysics Competition (IAAC) Gold Medalist

International Youth Math Challenge

International Youth Math Challenge (IYMC) Bronze Medalist

Social Impact & Activities

SPADES

Organizer

Organized a month-long initiative teaching coding and mathematics to 100+ underprivileged students.

EngagementVOLUNTEER
Environmental Organizations

Collaborator

Collaborated with environmental organizations to distribute smog detectors across 10 schools in Lahore, raising air-pollution awareness.

EngagementVOLUNTEER
SPADES EventOTHER

NASA Engineer Talk: James Webb Space Telescope

Hosted Pakistan’s first NASA engineer talk, highlighting the James Webb Space Telescope project to 300+ attendees.

Skills & Interests

Pruning

Quantization

QAT/PTQ

Python

PyTorch

Model compression

Profiling & Benchmarking

Knowledge distillation

Git

Actor–critic RL

C

scikit-learn

C++

RAG

Communication

Working across multiple languages, enabling global collaboration and clear technical outreach.

English

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.