Skip to content
View CallSohail's full-sized avatar
🎯
Focusing
🎯
Focusing

Block or report CallSohail

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
CallSohail/README.md

Muhammad Sohail

Data & AI Engineer · Université d'Évry Paris-Saclay · Incoming PhD on sovereign and secure AI

Website Open WebUI community Hugging Face X


I build AI platforms that run on our own hardware. At the university's IT department I look after EveGPT, a self-hosted assistant for staff built on Open WebUI, and its student counterpart. Everything from GPU serving to retrieval to the tools people click on happens on-premise, with no data leaving campus.

In October 2026 I start a PhD on sovereign and secure AI at Université Paris-Saclay (IBISC lab). The research picks up where the day job leaves off: what happens to security when institutions run their own LLM pipelines, and how to spot prompt injection, abuse and data leaks in the logs.

What I work on

  • Self-hosted LLM serving with vLLM on multi-GPU servers (Qwen, GPT-OSS, embedding and reranker models)
  • RAG pipelines with RAGFlow, Docling and hybrid search plus reranking
  • Open WebUI extensions: tools, filters, pipes, actions and events
  • MCP servers that connect assistants to real institutional systems (wikis, ticketing, directories, search)
  • Speech-to-text for accessibility, including lecture and video transcription
  • Observability for AI services with OpenTelemetry, Prometheus and Grafana

Featured projects

Project What it is
openwebu-work Production-tested Open WebUI extensions: Study Mode, Quick Actions, a RAGFlow connector, onboarding events and more. MIT licensed.
sway-laya A word game played against the open-source Laya decision model. Live on Hugging Face Spaces.
werewolf-laya Werewolf where every villager is a Laya agent.
SEJOUR A French residence-permit eligibility simulator with a naturalisation calculator.
MultiModel-MediScan Multimodal medical scan analysis.
Text-Based Emotion Detection Research on detecting perceived emotion in text.

Stack

Python PyTorch vLLM LangGraph FastAPI Docker PostgreSQL Redis Grafana n8n Linux Astro

Background

  • MSc in Data Science and Network Intelligence, Télécom SudParis (Institut Polytechnique de Paris), 2025
  • BSc in Computer Science, University of Malakand
  • Freelance AI and data consulting, delivered remotely in English

Get in touch

If you run AI inside a university or public body and care about keeping data in-house, I'd like to hear how you do it. Reach me through callsohail.dev or open an issue on any repo.

Pinned Loading

  1. openwebu-work openwebu-work Public

    Production-ready Open WebUI extensions, including Tools, Filters, Pipes, Actions, and Events for agent workflows, RAG, learning, automation, and model enhancement.

    Python 34 3

  2. Bridging-the-Gap-in-Text-Based-Emotion-Detection Bridging-the-Gap-in-Text-Based-Emotion-Detection Public

    This task focuses on detecting perceived emotions in text, including predicting emotions (joy, sadness, fear, anger, surprise, disgust), estimating emotion intensity, and cross-lingual emotion dete…

    Jupyter Notebook 1

  3. MultiModel-MediScan MultiModel-MediScan Public

    Multimodel medican scan

    Python 1

  4. SEJOUR SEJOUR Public

    HTML

  5. sway-laya sway-laya Public

    A word game against the Laya decision engine

    Python

  6. werewolf-laya werewolf-laya Public

    A werewolf game where every villager is the Laya decision model

    Python