— ALI JAVANI

AI Researcher & Software Engineer.

Currently pursuing my M.Sc. in Artificial Intelligence (IASD) at Université Paris Dauphine - PSL. Specializing in Large Language Models, Reinforcement Learning, and high-performance Machine Learning systems.

Latest Work: Quantized ML on radiation-hardened FPGAs for High Energy Physics at CERN.

Experience & Research

Research Intern (Quantized ML on FPGA)

CERN Geneva, Switzerland

July 2025 — Sept 2025
  • Designed a mixed-precision Autoencoder for the LHCb PicoCal calorimeter, compressing signal pulses (32 samples → 2 latents).
  • Implemented Quantization Aware Training (QAT) using QKeras and hls4ml, deploying on radiation-hardened FPGAs with O(10) ns inference latency.
  • Diagnosed 'Dying ReLU' convergence failures, optimizing architecture for a 4x reduction in size without accuracy loss.

Research Assistant (ML in Finance)

RiskLab - University of Toronto Remote

Dec 2023 — Jun 2024
  • Explored stability selection and ensemble feature importance methodologies for robust financial ML feature engineering.
  • Assessed novel VIX-driven conditional correlations via GRF to reduce hedge fund volatility versus standard factor models.
  • Contributed to the RiskLabAI (PyPI) open-source package by implementing Lopez de Prado's ML techniques.

Quantitative Researcher & Co-Founder

Wave Technologies Isfahan, Iran

Sept 2021 — Nov 2022
  • Designed a distributed PyTorch training platform featuring logging, crash recovery, and hardware-efficient parallel training.
  • Implemented LSTM, Autoencoder, and Transformer-based models for dynamic and static financial portfolio optimization.
  • Built a Python-MQL bridge for seamless model integration, enabling rapid testing and production with caching acceleration.

Back-end Developer Intern

Divar Tehran, Iran

Jun 2021 — Sept 2021
  • Implemented Kubernetes and Docker for scalable deployments and isolated development environment setups.
  • Integrated Redis and Celery for asynchronous task management and set up CI/CD pipelines with comprehensive unit testing.

Technical Arsenal

A comprehensive overview of the frameworks, languages, and infrastructure tools I use to build autonomous agents and robust engineering systems.

Machine Learning & Data

PyTorchPyTorch LightningTensorFlowKerasQKerasNumPyPandasScikit-LearnOpenCVWandBSpark

MLOps & DevOps

KubernetesDockerCI/CDGit

Programming Languages

C/C++PythonMQL4/MQL5JavaJavaScriptRScala

Databases

MySQLSQLitePostgreSQLMongoDBRedis

Web Development

DjangoReact.jsHTML/CSS/Bootstrap/MUI

CAD Tools & HDLs

VerilogQuartusModelSim

Featured Architecture

A selection of my academic research, full-stack applications, and experimental AI agents.

WeCare Recommender Systems

Ranked #1 among M2 Master 2025 groups. Implemented and ensembled advanced matrix completion models including Neural CF (NeuMF), Deep-MF, bias-aware ALS, and Graph Neural Networks (GCN/GAT).

PyTorchGNNsNeuMFALSCollaborative Filtering

Nucleus Sampling LLM Engine

Replication of 'The Curious Case of Neural Text Degeneration'. Built a custom generation engine using Hugging Face Transformers to evaluate top-p, top-k, and beam search decoding strategies on Qwen and GPT-2 models using KV caching.

PythonHugging FaceLLMsTransformersKV Caching

Agentic Zork

An autonomous Large Language Model agent that uses the ReAct methodology to map out, navigate, and solve classic text-based adventure games.

PythonLangChainLLMsReinforcement Learning