Open for collaboration · Paris, France

Data & AI
Engineer building
production systems. Muhammad Sohail — turning research into reliable AI infrastructure for institutions and teams.

I design and ship enterprise-grade AI: RAG platforms, LLM serving pipelines, and intelligent assistants used daily by hundreds of people. Currently engineering AI at Université d'Évry Paris-Saclay — moving the work from notebooks to dependable production.

Start a project View selected work
DATA · AI · ENGINEER
Available
for hire
2+
Years building AI
10+
Production systems
500+
Daily active users
4
Languages spoken
01 — About

An engineer who ships,
not just prototypes.

I'm Muhammad Sohail, a Data & AI Engineer working at the intersection of applied research and production engineering. My focus is making large language models useful and reliable for real organizations — not just impressive in a demo.

Originally from Pakistan, I now live and work in Paris, where I hold an M.Sc. in Data Science & Network Intelligence from Télécom SudParis at Institut Polytechnique de Paris. I currently work as a Data & AI Engineer at Université d'Évry Paris-Saclay, where I lead the design and deployment of the institution's internal AI platform.

My work spans the full stack of modern AI engineering: retrieval-augmented generation, LLM serving and orchestration, containerized deployment, observability, and the careful integration work that turns raw models into tools people actually want to use. I care about the parts most demos skip — auth, monitoring, multilingual UX, latency budgets, graceful failure.

Outside the office I'm a curious reader, an occasional cook, and I'm slowly learning French. I'm always open to interesting collaborations, freelance work, and conversations about AI systems that need to actually work.

— Muhammad
02 — Experience

Where I've shipped.

Roles, responsibilities, and the work that actually made it to production.

Sep 2025
— Present Current

Data & AI Engineer

Université d'Évry Paris-Saclay · Évry, France

Leading the design and deployment of the university's internal AI platform, serving hundreds of staff daily across HR and IT departments. My work focuses on end-to-end reliability — from model serving infrastructure to user-facing chatbots, observability, and integration with existing institutional systems.

  • Architect and maintain a production RAG platform powering everyday assistance for academic and administrative staff.
  • Build multimodal document pipelines that extract structured knowledge from complex PDFs (tables, figures, scanned forms).
  • Operate self-hosted LLM serving with hybrid search, reranking, and multilingual (FR/EN) support.
  • Develop custom integrations: directory search, ticketing, transcription, and workflow automation tools.
  • Set up monitoring, logging, and observability so the platform is debuggable when something inevitably breaks.
RAG LLM Serving OpenWebUI Docker PostgreSQL Observability Python FastAPI
Mar 2025
— Jun 2025

Data Science Intern — Master's Thesis

Université d'Évry Paris-Saclay · IBISC Laboratory, IRA2 Team

Built the first production version of the institutional HR & IT chatbot during my master's thesis internship — the prototype that became the platform I now maintain. Focused on validation, accuracy, and policy-compliant responses using exclusively open-source models for data sovereignty.

  • Designed a Retrieval-Augmented Generation pipeline for HR queries (recruitment, leave, training).
  • Engineered document parsing using Unstructured.io, Docling, LlamaParse, pdfplumber, and LangChain.
  • Delivered a working internal-facing app deployed within the university data center.
  • Thesis evaluated positively by the academic jury.
LangChain Streamlit FAISS Mistral Llama Python
Oct 2023
— Apr 2024

Data Analyst Intern

CodeClause Pvt Ltd. · Remote

Worked across data preparation, statistical modeling, and reporting. Cleaned and standardized messy datasets, ran regression and hypothesis testing, and produced dashboards that informed business decisions.

  • Built ETL routines for messy real-world data (missing values, inconsistent formats).
  • Delivered interactive dashboards and reports for stakeholders.
  • Practiced communicating statistical findings clearly to non-technical audiences.
Python Pandas SQL Power BI Scikit-learn
03 — Education

From Malakand to Paris.

A path shaped by curiosity, scholarship, and the belief that good engineering starts with good fundamentals.

Sep 2024 — Sep 2025

M.Sc. — Data Science & Network Intelligence

Télécom SudParis — Institut Polytechnique de Paris, France
GPA 15.81 / 20 · Mention
Data Science Deep Learning Network Intelligence Distributed Systems IoT Data Visualization Blockchain Business Modeling
Aug 2019 — Sep 2023

B.Sc. — Computer Science

University of Malakand, Pakistan
GPA 3.75 / 4.0 · First Class
Programming Algorithms & DS Databases Operating Systems Networking Software Engineering Artificial Intelligence Mathematics & Statistics
04 — Selected work

Projects that made it
to real users.

A selection of production systems, applied research, and side projects I've built over the past two years.

01

Institutional RAG Platform & AI Assistant

End-to-end retrieval-augmented chat platform serving university staff for HR and IT queries. Handles multimodal document ingestion (PDFs with tables, scanned forms, images), hybrid semantic + lexical search, multilingual responses (French & English), source attribution, and role-based access. Iterated through multiple versions with continuous user feedback — moved from a single-server prototype to a multi-component platform with proper auth, monitoring, and graceful degradation.

PythonLangChainOpenWebUIRAGFlowFAISSpgvectorDockerNginxFastAPI
Live · 2025
02

Meeting Transcription & Summarization Pipeline

Automated transcription system for video conferencing infrastructure. Triggers on recording completion, runs Whisper-based ASR with speaker diarization for French and English, and uses LLM summarization to produce structured meeting notes with action items. Tuned for accuracy on multilingual academic discussions, with graceful handling of partial failures.

PythonWhisperXFaster-WhisperLangChainBashDocker
Live · 2025
03

Custom AI Tools & Integration Layer

A growing library of internal tools that extend the chatbot platform — directory lookups, wiki search, ticketing automation, email helpers, and a code interpreter. Built as modular Python pipes and filters with strict permission boundaries, designed so each new tool can be added without destabilizing existing ones.

PythonFastAPIPydanticLangGraphn8nSOAP/REST APIs
Live · 2025
04

CV Screening & Recruitment Agent

Multi-agent recruitment pipeline using LangGraph that screens CVs, extracts structured profile data, scores against job requirements, and drafts initial interview questions. Designed as an explainable system — every decision links back to the specific evidence in the source document.

PythonLangGraphLangChainOpenAI APIPydantic
Shipped · 2024
05

MediBot — Medical Information Chatbot

A health-information chatbot retrieving trusted medical content via context-based RAG retrieval. Implemented FAISS vector store with Mistral-7B, response source-attribution, and a clean Streamlit interface. Built as a learning exercise in domain-specific RAG with careful attention to safety disclaimers and source transparency.

PythonLangChainFAISSMistral-7BHugging FaceStreamlit
Shipped · 2024
06

Parkinson's Disease Progression Prediction

Applied machine learning to identify Parkinson's disease subtypes and predict progression from clinical and genetic features. Used gradient-boosted ensembles with thoughtful feature engineering and built an interactive Streamlit dashboard to help clinicians explore patterns and model outputs.

PythonXGBoostLightGBMScikit-learnPandasStreamlit
Research · 2023
05 — Toolkit

The tools I reach for.

Technologies I use day-to-day to design, build, and ship AI systems that hold up under real load.

Languages
Python SQL JavaScript Bash C++ Java HTML/CSS
AI & LLMs
RAG LangChain LangGraph LlamaIndex Hugging Face vLLM Ollama OpenAI API Mistral Anthropic API Prompt Engineering Fine-tuning
Machine Learning
PyTorch TensorFlow Scikit-learn XGBoost LightGBM Keras Pandas NumPy Feature Engineering Model Eval
Data & Storage
PostgreSQL MySQL pgvector FAISS ChromaDB Redis SQLite ETL Pipelines
Infrastructure & DevOps
Docker Docker Compose Linux Nginx Git / GitHub GitLab CI/CD NVIDIA CUDA REST APIs OpenTelemetry Grafana
Frameworks & UI
FastAPI Streamlit Gradio OpenWebUI RAGFlow n8n Power BI Tableau Matplotlib Seaborn
06 — Principles

How I think
about AI engineering.

Four convictions that shape every project I take on.

— 01
Ship, don't demo.
A model on a laptop isn't a product. Real value comes from systems that handle auth, monitoring, failure modes, and the boring parts most demos skip.
— 02
Sovereignty matters.
For institutions and regulated industries, open-source self-hosted models are often the right answer. Data ownership is a feature, not a constraint.
— 03
Boring is reliable.
Postgres, Docker, clean APIs, good logging. Most production problems are solved by mature, unglamorous tools — not the newest framework.
— 04
Measure or guess.
Retrieval quality, latency, cost, accuracy on real queries — if you're not measuring it, you're guessing. Evaluation is part of the build.
07 — Research

Applied research bridging
theory and practice.

My thesis work and side research, focused on making advanced AI techniques work in real institutional contexts.

2025
Master's Thesis

Intelligent Multimodal RAG for University Information Management

Télécom SudParis · IBISC Laboratory, IRA2 Team

End-to-end Retrieval-Augmented Generation system for automating university HR and IT support. Explored multimodal document parsing, hybrid retrieval (semantic + lexical), open-source LLM selection for institutional compliance, and an enterprise readiness framework spanning freshness, architecture, cost, testing, and security. Deployed in production; positively evaluated by the academic jury.

2025
Applied Research

Building Production-Grade RAG with the FACTS Framework

Independent applied research

Practical implementation of the FACTS framework for enterprise RAG chatbots. Investigated advanced chunking, query rephrasing, multi-step retrieval, and guardrails — using exclusively open-source models to preserve data privacy within institutional infrastructure.

2023
Academic Project

Parkinson's Progression Prediction with Gradient Boosting

University of Malakand · Final-year project

Machine-learning system to identify Parkinson's disease subtypes and predict progression from clinical and genetic features. Used ensemble methods (XGBoost, LightGBM) with careful feature engineering, paired with an interactive Streamlit dashboard for clinicians.

08 — Services

How I can help.

Available for freelance projects, consulting engagements, and longer-term collaborations.

— 01

RAG & AI Chatbots

Production retrieval-augmented systems for support, knowledge management, and internal automation — built with open-source LLMs when data privacy matters.

  • Document ingestion & processing
  • Hybrid retrieval & reranking
  • Multilingual response handling
  • Evaluation harnesses
— 02

AI Infrastructure

Self-hosted LLM serving with vLLM or Ollama, containerized deployment, monitoring, and the integration glue that connects models to existing systems.

  • Docker / Compose deployments
  • GPU-optimized serving
  • OpenTelemetry & Grafana
  • Nginx, auth, hardening
— 03

Data & Analytics

Interactive dashboards, ETL pipelines, and automated reporting. From messy raw data to clean insights stakeholders actually read.

  • Pandas / SQL pipelines
  • Power BI & Tableau dashboards
  • Statistical reporting
  • Predictive models
— 04

Custom Integrations

API integrations, custom tools, AI workflow automation. From transcription pipelines to multi-agent systems orchestrated with LangGraph.

  • FastAPI services
  • LangChain / LangGraph agents
  • Workflow automation (n8n)
  • Third-party API integration
09 — Get in touch

Let's build
something that lasts.

Open to collaboration, consulting, and conversations about AI systems that need to actually work in production. Replies usually within 24 hours.

sohail.cs951@gmail.com
Évry, Paris region, France
+33 7 81 58 87 98
Responding within 24h