Areeb Abad Fazli

Areeb Abad Fazli

ML Engineer · RAG & agentic systems

Barcelona

I build retrieval and agentic systems and evaluate them against labelled data. When a gain isn't statistically significant, I don't claim it.

  1. Retrieval systemshybrid BM25 + dense · RRF fusion · rerankers
  2. Evaluationlabelled sets · paired significance tests · ablations
  3. Agentic systemsLLM orchestration · PR review pipeline · LLM query layer

About

I'm an ML engineer and AI researcher specialising in RAG pipelines and agentic systems. I completed an MSc in Intelligent Interactive Systems at Pompeu Fabra University in Barcelona (2024–2026), where my thesis built a neuro-symbolic bridge that explains a chess engine's reasoning in natural language. Before that, I built client websites in Python (Django, Flask) and JavaScript as a backend developer.

Selected projects

RepoSentinel

An AI security reviewer for pull requests: an LLM audits the whole change, a separate verifier call double-checks every finding, and a GitHub Action posts the confirmed ones as inline comments on the exact line.

  • Python
  • FastAPI
  • Semgrep
  • tree-sitter
  • Qdrant
  • GitHub Actions
  • Similarity search can't tell a bug from its fix (precision 0.5 at every threshold on 1,054 functions), so retrieval is only context for the LLM review.
  • Static checks alone reach 23% of bugs at a 4% false-alarm rate (502 held-out bug/fix pairs), so they serve as leads, not verdicts.

Other projects

  • Football Player Performance Prediction: a Flask REST backend serving regression predictions, with a Streamlit dashboard for model outputs, feature importances and validation metrics.
  • Feedifying: a full-stack aggregator that consolidates technology news from multiple sources into a single feed (Flask, PostgreSQL).

Experience

Backend Developer

SigmaX Marketing · Aligarh, India

  • Developed and shipped full-stack web applications for client businesses in Django, Flask and Node.js with React front ends, owning most projects from requirements through delivery, using AWS on some.
  • Diagnosed and fixed slow pages on client sites by profiling requests and eliminating N+1 query patterns and ORM misconfiguration.
  • Refactored data access from ad-hoc SQL to SQLAlchemy ORM on PostgreSQL, leaving a cleaner, testable data layer.

Geospatial Data Science Intern

Aligarh Muslim University · Aligarh, India

  • Built Python pipelines (Pandas, NumPy) to preprocess large-scale geospatial datasets, engineer ML-ready features and visualise spatial patterns.
  • Analysed Waqf board land data across India, cross-verifying ground survey readings against satellite-derived data.

Education

  • MSc Intelligent Interactive Systems, Pompeu Fabra University, Barcelona · 2024–2026
  • Bachelor of Computer Applications, Jamia Hamdard, New Delhi · 2019–2022

Publication

Award

  • Winner, PLYO Lab Hackathon, Barcelona · 2025. Built Waldo AI in a team of three in under 48 hours: a tool that maps employee connections across VC portfolio companies to surface warm introductions. Organiser's post

Skills

AI/ML

  • RAG pipelines
  • Hybrid retrieval (BM25 + dense)
  • LLM orchestration
  • Qdrant
  • Hugging Face
  • PyTorch
  • TensorFlow
  • Scikit-Learn
  • Explainable AI
  • Neuro-symbolic AI
  • MLflow
  • DVC

Research & evaluation

  • Retrieval evaluation on BEIR (nDCG@k, Recall@k, MRR, MAP)
  • ranx
  • Paired significance testing
  • Stage-by-stage pipeline ablations
  • Reproducible experiment harnesses

Languages & frameworks

  • Python
  • C
  • TypeScript
  • JavaScript
  • FastAPI
  • Flask
  • Django
  • React
  • Node.js
  • SQL (PostgreSQL, MySQL)
  • MongoDB
  • Git

Cloud, DevOps & testing

  • AWS (S3, EC2, SageMaker)
  • Docker
  • Kubernetes
  • CI/CD
  • GitHub Actions
  • pytest
  • Datadog
  • Grafana

Contact

If you'd like to get in touch, email is the best way to reach me.