About
Data Scientist with real-world experience in deep learning, NLP, and computer vision. Trained voice cloning models, built multi-sensor activity recognition pipelines, and co-developed an AI-powered education platform end-to-end. Fluent in Python and SQL, with a solid grasp of the full model lifecycle from data preprocessing to model deployment. Currently deepening expertise in large language models. Seeking a data-driven role where I can build impactful ML solutions from day one.
Skills & Expertise (21)
Work Experience
Data Science Intern
Meril Life Science
Dec 2025 - Present
Reduced PostgreSQL query response times by ~40% by redesigning table schemas, adding composite indexes on high-frequency join columns, and rewriting 10+ slow queries — resolving the primary bottleneck in a production data environment. Gained working knowledge of AWS cloud infrastructure by configuring S3 for object storage and EC2 for compute workloads, supporting production ML pipeline deployment at scale. Engineered RoboMinds, an AI-powered coding education platform for Grade 9-12 students, shipping 10 progressive maze levels, a BFS-based AI hint engine, and a LiveKit voice agent — deployed on Netlify and Render for live public access. Crafted 10+ real-time bone animations (walk, wave, dance, jump, crawl) using Three.js AnimationMixer and 5 custom Blockly block categories, enabling students to control robot behaviour through visual block-based code.
Data Science Intern
PetPooja
Nov 2024 - May 2025
Built a few-shot voice cloning system by fine-tuning the XTTS deep learning model on under 5 minutes of clean audio per speaker, achieving 0.83–0.91 cosine similarity across 8 speakers. Deployed the voice cloning model on Hugging Face Spaces via Gradio, delivering data-driven voice synthesis at under 5-second inference on a CPU-only environment with no GPU infrastructure required. Designed a sensor fusion ML pipeline that classified 18+ simultaneous human activity types at high F1 scores, by combining body-worn, object-based, and ambient data streams into a unified feature space. Boosted overlapping activity detection reliability by tuning multi-label decision thresholds across 18 label pairs, cutting misclassification on the most frequent co-occurring activity combinations. Produced 3 Power BI dashboards adopted for weekly cross-functional team reporting, converting raw restaurant transaction data into actionable visual summaries — after earning PetPooja’s internal Power BI certification.
Education
M.Sc. Big Data Analytics - St. Xavier’s College
2023 - 2025 · Afghanistan
Bachelor of Computer Applications - Lachoo Memorial College
2020 - 2023 · Afghanistan
Certifications
Power BI Training
PetPooja · 2025
IBM SkillsBuild Internship Program — AI/ML
IBM · 2024
IBM SkillsBuild Micro-Internship: Data Science
IBM · 2024
IBM SkillsBuild Micro-Internship: Generative AI
IBM · 2024
Tata Group Job Simulation: Data Visualization
Forage / Tata · 2024
Interested in this developer?
Profile Score Breakdown
Profile Overview
Skills (21)
Click a skill to find developers with the same skill
