Marwa Zulfiqar
LLM & Agents
Python
SQL
Model Architecture & Training: 5/10
Data Pipeline & Feature Engineering: 5/10
Experimentation & Evaluation: 5/10
Active 1 day ago
Invite to interview
Download CVCV
Overview
Technical skills
Projects
Timeline
Roles
Overview
ML engineer with end-to-end project experience at a senior level focusing on knowledge graph pipelines and embedding-based models. The strongest proven skill is knowledge-graph engineering and KGE training as shown by src/kge/train_eval.py and src/kg/expand.py which implement training, evaluation, SPARQL expansion and enrichment workflows. There is limited evidence of production MLOps, advanced efficiency engineering like quantization, and formal test coverage or deployment automation in public code.
Phone
Technical skills
Languages
2
Python
SQL
AI/ML
12
Pandas
Scikit-learn
NumPy
Streamlit
spaCy
LLM
Llama
Mistral
Anthropic
LIME
SHAP
Jupyter Notebook
DevOps
4
GCP
GitHub
Docker
Git
Other
10
Requests
Matplotlib
Beautiful Soup
Google BigQuery
AI Agents
Machine Learning
RAG
dbt
Knowledge Graph
Explainable AI
Projects
Explainable Bank Fraud Detection
Built an explainable machine learning model for fraud detection. Used SHAP and LIME to generate interpretable explanations for model outputs. Validated the approach so results can be reviewed and better controlled by people.
AI Agents for Portfolio Management
Developed AI agents intended to manage a portfolio autonomously. Used open-source LLMs and APIs to orchestrate agent behavior and workflows. The project is ongoing and currently focused on improving agent capabilities for decision support.
Knowledge Graph: Mapping, Classification & AI Search
Built a structured knowledge graph and an AI search interface over it. Implemented an LLM+RAG natural-language querying flow backed by a knowledge graph API. Created a Streamlit-based application to support user questions over multi-source data.
Timeline
Data Engineer Intern
•
Junior
Nurel.pk
•
Internship
Integrated data from multiple source systems into a reliable Python and SQL pipeline. Prepared, structured, and documented datasets for downstream users. Implemented data quality checks and technical standards to improve trust and usability across teams.
Python
SQL
Deep Learning Research Assistant
•
Middle
Prince Sattam University
•
Full-Time
Co-authored a peer-reviewed research article by testing and comparing different model architectures. Followed a rigorous, well-documented experimental methodology. Contributed to the evaluation of approaches for agricultural-related tasks.
Machine Learning Intern
•
Junior
EziTech Learning Institute
•
Internship
Cleaned and preprocessed datasets for machine learning training. Trained and evaluated ML models using Python with Pandas and scikit-learn. Focused on improving data readiness and model performance through systematic evaluation.
Python
Pandas
Scikit-learn
ML / Web Developer Intern
•
Junior
GrowOn (SMC-Private) Limited
•
Internship
Built predictive models and KPI dashboards for an e-commerce use case. Prepared and presented results for non-technical stakeholders. Supported analytics deliverables focused on business performance monitoring.
Middle AI/ML Engineer
Confidence: Medium ML Engineer
ML engineer with end-to-end project experience at a senior level focusing on knowledge graph pipelines and embedding-based models. The strongest proven skill is knowledge-graph engineering and KGE training as shown by src/kge/train_eval.py and src/kg/expand.py which implement training, evaluation, SPARQL expansion and enrichment workflows. There is limited evidence of production MLOps, advanced efficiency engineering like quantization, and formal test coverage or deployment automation in public code.
Model Architecture & Training
5/10
How well models are designed and trained
Solid applied model training and evaluation for knowledge graph embeddings using PyKEEN and explicit training/evaluation flows; uses established libraries rather than fully custom neural architectures.
Data Pipeline & Feature Engineering
5/10
How data is prepared for models
End-to-end data acquisition and cleaning pipeline with crawler, deduplication, normalization, corpus building and merging of KG artifacts.
Experimentation & Evaluation
5/10
How results are measured and tested
Reasonable experimentation and evaluation practices including held-out test evaluation, rank-based metrics for KGE, size-sensitivity experiments, t-SNE visualizations and grid search in notebooks.
MLOps & Deployment
3/10
How models are shipped to production
Basic model persistence and a Streamlit UI for interaction; limited evidence of production-grade MLOps like CI/CD, monitoring, model versioning or serving infrastructure.
Computational Efficiency
2/10
How efficiently computing resources are used
Some attention to efficiency in visualization and experiments (subsampling for t-SNE, batching), but no advanced GPU optimization, quantization or profiling artifacts.
Research Depth & Innovation
4/10
Depth of research and new ideas
Applied research-level thinking around KG expansion, entity alignment and relation arithmetic, but no novel algorithms or formal research artifacts reproducing SOTA.
Verified artifacts
Expertise
RAG• Middle
Graph Machine Learning (GNN)• Middle
Industries
Farming & Agriculture• Middle
Lifestyle• Middle
Sports• Middle
Technologies
SQL• since 2026 • Junior
GCP
Jupyter Notebook
SHAP
dbt
AI Agents
Llama
Mistral
LIME
Git
Docker
LLM
RAG
Google BigQuery
Explainable AI
Anthropic
GitHub
Knowledge Graph
Machine Learning• since 2026
Knowledge Graph• mentioned only
NER• mentioned only
NLP• mentioned only
RAG• mentioned only
Recommendations
- Develop RAG-backed applications and KG-backed QA systems that combine entity alignment, SPARQL generation and LLM prompting.
- Implement productionization work for existing pipelines: model versioning, CI/CD, containerized serving and lightweight monitoring for the Streamlit apps and KGE models.
- Focus on efficiency engineering for KGE training and inference such as batching, mixed precision, profiling and optionally integration with optimized runtimes if low-latency serving is required.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
Middle Data Scientist
Confidence: Medium Data Engineer
Data engineer at a senior level specializing in knowledge-graph pipelines and data engineering for web-sourced information. The strongest proven skill is building end-to-end ingestion and KG enrichment pipelines as demonstrated by the crawler, cleaner and KG expansion code (src/crawl/crawler.py and src/kg/expand.py). There is limited evidence of production MLOps, automated tests, distributed big-data systems, or formal data governance in public code.
Statistical Rigor
5/10
Correct use of statistics
Moderate statistical rigor: explicit evaluation metrics, experiments and cross-validation are present, but full statistical treatment (uncertainty quantification, multiple comparisons, causal analysis) is limited.
Data Wrangling & Cleaning
7/10
Preparing and cleaning data
Strong data wrangling and cleaning practices: careful crawler with polite rate limiting, recursive cleaning, deduplication and date normalization, and graph cleaning for KG expansion.
Evidence
WeiS49/premier-league-knowledge-graph/src/crawl/crawler.py:fetch_page, save_raw_html, extract_infobox and get_player_links
WeiS49/premier-league-knowledge-graph/src/crawl/cleaner.py:clean_dict, normalize_date, deduplicate, clean_teams and clean_players
WeiS49/premier-league-knowledge-graph/src/kg/expand.py:clean_graph and enrich_new_entities (graph-level cleaning and filtering)
Exploratory Analysis & Visualization
6/10
Exploring and visualizing data
Exploratory analysis and visualization are practiced and integrated into outputs (Streamlit dashboard, t-SNE and feature importance plots), though narrative interpretation is uneven across projects.
Evidence
marwa525/Paris-Fashion-Intelligence/app.py:plotly visualizations and Streamlit presentation of insights
WeiS49/premier-league-knowledge-graph/src/kge/train_eval.py:tsne_visualization and summary printing
Ash12341/Plant-Disease-Detection_ML-Project-DIA3/Plant-Disease-Detection_ML-Project-DIA3.ipynb:EDA, correlation heatmap and interpretation cells
Predictive Modeling
6/10
Building models that predict
Predictive modeling competency is solid: baseline tuning, cross-validation, KGE training and evaluation pipelines exist; some experiments and error analysis are present though full production readiness (calibration, advanced error analysis) is not evidenced.
Business Insight & Impact
3/10
Turning analysis into business value
Some product-minded presentation is present (a Streamlit dashboard and short conclusions), but concrete business-metric framing, cost-of-error analysis, or deployment impact are scarce.
Reproducibility & Notebook Hygiene
4/10
Clean, repeatable analysis
Reproducibility practices appear partially addressed via requirements files, argparse entry points and fixed random seeds in training code, but there is no evidence of CI, DVC/data versioning, or extensive notebook-to-module refactoring.
Evidence
marwa525/Paris-Fashion-Intelligence/requirements.txt:requirements pinned
WeiS49/premier-league-knowledge-graph/src/kge/train_eval.py:argparse entrypoint and random_seed usage
Ash12341/Plant-Disease-Detection_ML-Project-DIA3/Plant-Disease-Detection_ML-Project-DIA3.ipynb:# NOTEBOOK META (executed notebook with reasoning cells)
Expertise
Data Science• Middle
Industries
Farming & Agriculture• Middle
Sales & Marketing• Middle
Sports• Middle
Technologies
Scikit-learn• since 2024
Matplotlib
spaCy
Pandas• since 2024
NumPy
Streamlit
Knowledge Graph• mentioned only
NER• mentioned only
NLP• mentioned only
RAG• mentioned only
Recommendations
- Develop production-ready ETL and KG pipelines for structured knowledge ingestion, entity alignment and Wikidata expansion (crawler + expand modules).
- Build interactive analytics dashboards and lightweight monitoring for model and pipeline outputs using Streamlit and plotting (app.py style).
- Lead experiments on knowledge-graph embeddings and evaluation (PyKEEN training and nearest-neighbor/tsne analysis).
- Harden reproducibility and deployment: add tests, CI, data versioning (DVC), and containerized deployment for models and pipelines.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
Middle Backend Developer
Confidence: Medium Data Platform
A data-platform-focused Python engineer at a middle level with a strength for building end-to-end knowledge graph and web-data pipelines. The strongest proven skill is knowledge-graph construction and enrichment, evidenced by the expand.py alignment/expansion logic and the KGE training/evaluation pipeline (src/kg/expand.py and src/kge/train_eval.py). What is not evidenced is production-grade backend engineering such as database transactions/migrations, service-level observability, formal auth/secret management, or automated tests.
API Design
2/10
How well APIs are designed
Basic API usage and integration present but no deliberate API contract design, versioning or idempotency strategy; evidence is limited to external API calls and simple heuristics.
Data Layer & Database
3/10
Working with databases
Reasonable data-layer scripts and KG handling with cleaning, deduplication, and RDF graph manipulation, but no transactional DB design, migrations, or explicit isolation control.
Evidence
WeiS49/premier-league-knowledge-graph/src/crawl/cleaner.py: clean_teams, clean_players, deduplicate and normalize_date functions for schema hygiene
WeiS49/premier-league-knowledge-graph/src/kg/expand.py: building and serializing expanded RDF graph (expanded.nt) and compute_stats
WeiS49/premier-league-knowledge-graph/src/kge/train_eval.py: make_triples_factory and TriplesFactory usage for KGE data preparation
Scalability & Performance
3/10
Handling load and speed
Some scaling and performance awareness (batching, rate limits, subsampling) but no measured load testing, sophisticated caching strategy, or connection pooling setup.
System Architecture
3/10
Overall system structure
Clear modular decomposition into crawl/ie/kg/kge/rag/reason components showing intentional separation of concerns, but this is a single-repo student-style pipeline without production service contracts or multi-service resilience patterns.
Evidence
Project layout: src/crawl, src/ie, src/kg, src/kge, src/rag, src/reason directories (multiple files across these modules)
WeiS49/premier-league-knowledge-graph/src/kg/expand.py: orchestrates multiple expansion strategies and merging with original KG
Security & Auth
2/10
Protecting data and access
Minimal security posture: some safe defaults like request timeouts and User-Agent headers, but insecure handling of API keys in query strings and no auth lifecycle/secret management or input sanitization patterns.
Reliability & Observability
3/10
Stability and monitoring
Basic reliability practices present: retries with backoff for SPARQL, try/except around training and fetching, and polite delays; lacking structured logging, metrics, circuit breakers and formal graceful shutdown flows.
Expertise
Python• Middle
Databases & Vector Storage• Middle
Backend AI & LLM• Middle
Industries
Sales & Marketing• Middle
Sports• Middle
Technologies
Python• since 2024 • Middle
Beautiful Soup
Requests
Recommendations
- Develop knowledge-graph ETL and enrichment pipelines that ingest web data, align to external IDs, and materialize RDF artifacts (teams, entities, expansions).
- Build data-cleaning and NER preprocessing tooling for web-scraped content and maintain corpus pipelines for downstream ML models.
- Prototype RAG/LLM-backed analytics (short-lived experimental systems) and KGE experiments that need data engineering and evaluation scripts.
- Implement production hardening on existing pipelines: secrets management for API keys, structured logging, metrics and simple retry/backoff with jitter.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories: