Help & Supportcareers@gravityer.com
AI Engineer with 1+ years in Computer Vision & Machine Learning
AI is analyzing your overall score…
Identifying your key strengths…
Evaluating your skill match against the job requirements…
Assessing your cultural and operational fit
End-to-end AI/ML engineer who builds systems that close the gap between research and deployment. Six published works spanning graph neural networks, multimodal fusion, and ASR robustness, each grounded in real-world data and validated against production constraints. Designs architectures from first principles, resolves failure modes independently, and ships reproducible pipelines on local GPU hardware (RTX 4070 Super). Drawn to problems at the intersection of scientific rigour and engineering craft.
COMSATS University Islamabad
Bachelor of Artificial Intelligence · Artificial Intelligence
January 1, 2021 – June 1, 2026
ASR Robustness Research Pipeline
May 1, 2026 – June 1, 2026
Designed synthetic noise injection at 6 WER levels, QLORA fine-tuned Qwen2.5-3B on SLURP, and ran controlled ablations across three fully-trained models, producing two ACL ARR submissions with a full public reproducibility package.
View ProjectAMR Simulation Lab
March 1, 2026 – June 1, 2026
Built an end-to-end AMR research platform: Mesa 3.x agent-based bacterial simulation encoding SOS response, HGT conjugation, biofilm formation, persister switching, and Hill-equation pharmacodynamics, validated against 123,189 BV-BRC isolates. Trained 3-layer GAT on 956,446 directed edge-gene pairs from 312 simulation snapshots; exposed 15+ REST endpoints and WebSocket streams via FastAPI with a real-time HGT visualisation frontend.
View ProjectFaceFuel v3: Visual Health Intelligence System
March 1, 2026 – April 1, 2026
Assembled a 17,343-image dataset across 3 modalities; trained YOLO11m detectors, extracted DINOv2 ViT-S/14 features (3072-dim), and implemented Bayesian product-of-experts fusion, producing three peer-reviewed publications from a single system.
View ProjectHumanatic: Fully Automated Intelligent Call Classifier
January 1, 2025 – December 31, 2025
Self-sourced the entire training corpus over several months: collected approximately 20,000 outbound call recordings (5 seconds to 10 minutes each), transcribed using Whisper v3, and hand-labelled across a five-class automotive intent taxonomy (new appointment, prior sale follow-up, test drive, general enquiry, no-connect / wrong number). Benchmarked BERT and ROBERTa as baselines, then fine-tuned Qwen and DeepSeek LLMs on the labelled corpus with synthetic data augmentation; best model achieved approximately 98% classification accuracy. Built an end-to-end Selenium and BeautifulSoup automation pipeline that submits classifications to the Humanatic platform autonomously, replacing the entire manual review workflow at scale.
JARVIS: Autonomous AI Agent Framework
January 1, 2025 – December 31, 2025
Built a self-correcting autonomous agent for code generation, modification, and testing via planning, execution, and visual verification loops using Qwen2.5-VL, fully local with no cloud dependencies.
AMRResistanceGNN: Predicting Horizontal Gene Transfer of AMR Genes via Graph Attention Networks
IEEE Journal of Biomedical and Health Informatics (Under Review)
June 1, 2026 – Present
How Robust Are Intent Classifiers to ASR Noise? TF-IDF vs BERT vs LLMs
Zenodo, ACL ARR (Under Review)
May 1, 2026 – Present
Noise-Augmented Training Cuts LLM Degradation Under ASR Errors by 47%
Zenodo, ACL ARR (Under Review)
May 1, 2026 – Present
Multi-Modal Fusion via Product-of-Experts (Bimodal Visual Health Assessment)
Zenodo, IEEE JBHI
April 1, 2026 – Present
FaceFuel: Multi-Stage Heterogeneous Fusion Pipeline for Skin Analysis
Zenodo, IEEE JBHI
April 1, 2026 – Present
TriModal: Face, Tongue, and Eye Deficiency Screening
Zenodo, IEEE JBHI
April 1, 2026 – Present
Cultural Fit Analysis
The candidate's diverse project portfolio, spanning AMR simulation, visual health intelligence, ASR robustness, and autonomous agents, indicates a broad interest in various AI applications. The focus on 'closing the gap between research and deployment' and 'scientific rigour and engineering craft' suggests a pragmatic and results-oriented approach, which aligns well with a culture that values both innovation and practical implementation. The independent researcher status and numerous publications also point to a self-motivated and proactive individual.
Soft Skills & Operational Fit
The candidate's profile highlights an ability to work independently, evidenced by self-sourced datasets and independent research projects. The emphasis on 'resolves failure modes independently' and 'ships reproducible pipelines' suggests a strong problem-solving aptitude and a focus on practical, deployable solutions. The numerous publications indicate a drive for rigorous validation and contribution to the field.