Help & Supportcareers@gravityer.com
AI Engineer with 2+ years in Machine Learning & Visual Token Compression
AI is analyzing your overall score…
Identifying your key strengths…
Evaluating your skill match against the job requirements…
Assessing your cultural and operational fit
Final-year CS undergraduate specializing in efficient multimodal learning, visual token compression, and KV cache compression for reasoning models. Experienced in developing and evaluating AI/ML pipelines at scale, mitigating class imbalance, and contributing to open-source projects. Proven ability in algorithmic analysis, proof techniques, and mentoring students across core CS and math courses.
Lahore University of Management Sciences (LUMS)
B.Sc. (Hons.) · Computer Science
N/A – June 30, 2027
Lahore University of Management Sciences (LUMS)
Research (What Accuracy Hides: Decomposing KV Cache Compression Failure in Reasoning Models)
June 1, 2026 – Present
India
Lahore University of Management Sciences (LUMS)
Research Experience (Adapter-Distribution Mismatch in VLM Compression Evaluation (Train Your Baselines))
January 1, 2026 – Present
India
Lahore University of Management Sciences (LUMS)
Research Experience (Visual Token Compression for Vision-Language Models (TC-ViTok))
January 1, 2025 – Present
India
Lahore University of Management Sciences (LUMS)
Undergraduate Teaching Assistant
January 1, 2024 – Present
India
EcoTrack - Android Sustainability App
January 1, 2026 – June 1, 2026
Shipped a gamified Android app on MVVM/Firebase computing real-time CO2/water/waste impact from EPA/IPCC emission factors, delivering 100% of scoped features in a 5-person team. Built QR check-in (CameraX + ML Kit), a geofenced anti-cheat system with rate limiting and anomaly detection, FCM push notifications, and offline-first Firestore sync.
View ProjectU.S. Accident Severity Predictor
September 1, 2025 – June 1, 2026
Built an end-to-end ML pipeline over 7.7M U.S. accident records, engineering temporal, weather, and weighted-daylight features to predict accident severity, achieving 54% recall on rare Severity 1 and 36% recall on Severity 4 (most dangerous) accidents. Mitigated ~80% class imbalance via cost-sensitive Random Forest weighting, reaching 83% accuracy and 95% majority-class recall on a held-out split; documented methodology in a public technical write-up.
View ProjectSOC Fraud Detection System
September 1, 2025 – June 1, 2026
Architected a 9-service Python business-logic layer (21 DTOS, 25+ custom exceptions) over SQL Server, with role-based access control, session-token auth, and account-lockout enforcement. Implemented an 8-step real-time fraud pipeline - velocity limits, SQL-UDF risk scoring, automated over-$10k alerts - feeding a 4-tier risk engine with full audit-trail, KYC, and case-lifecycle workflows.
View ProjectTrain Your Baselines: Adapter–Distribution Mismatch in Visual Token Compression Evaluation.
Asian Conference on Computer Vision (ACCV)
January 1, 2026 – Present
Why Jointly Learned Pre-LLM Visual Token Selectors Can Fail: A Diagnostic Case Study.
International Conference on Neural Information Processing (ICONIP)
January 1, 2026 – Present
Cultural Fit Analysis
The candidate's academic background at LUMS, coupled with multiple research roles and a TA position, demonstrates a strong commitment to learning and collaboration within an academic setting. The diversity of projects, from ML pipelines to Android app development and backend systems, indicates adaptability and a broad interest in different technical domains. The focus on rigorous evaluation protocols and open-source contributions suggests a value for transparency and community contribution. The candidate's profile aligns with a culture that values continuous learning, innovation, and a data-driven approach to problem-solving.
Soft Skills & Operational Fit
The candidate's extensive research experience, including first-author publications and ongoing projects, indicates strong problem-solving, critical thinking, and analytical skills. Their role as a Teaching Assistant suggests good communication and mentoring abilities. The detailed project descriptions highlight a methodical approach to complex problems and a commitment to reproducibility (e.g., releasing training code). The academic background and research focus align well with the demands of an AI Engineer role, particularly one involving cutting-edge research and development.