Skip to content
View MansurPro's full-sized avatar
😎
Working from Lab
😎
Working from Lab

Block or report MansurPro

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. 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
MansurPro/README.md

Mansurbek Satarov — AI Researcher, AI/ML Engineer, Full-Stack Engineer and Team Lead. The stack I work across: interfaces, services and APIs, agents and RAG, models and training, systems and infrastructure.

studying when AI systems can be trusted

Website Projects Research Contact


About

I'm an AI researcher who ships production software. As a fully funded Computer Science Ph.D. student at the University of Cincinnati (advised by Prof. Justin Zhan), I study when AI systems can be trusted, and I build the systems and evaluations that check it.

Research Engineering
  • LLM & agent security — how models and the agents built on them are attacked and defended
  • Privacy in ML — what private and collaborative training actually protect, and for whom
  • Alignment & training dynamics — how behavior forms during training and fine-tuning
  • Safety evaluation — building the evaluations that decide whether a model is safe to ship
  • At scale — experiments on open-weight models and tool-using agents on GPU clusters (Slurm, vLLM, DeepSpeed), reported with preregistered, held-out evaluation

Manuscripts submitted and in preparation at top ML and security venues.

  • Agents & RAG — LangGraph workflows, tool calling, MCP, grounded retrieval
  • Computer vision & serving — YOLO and ViT-Pose inference, GPU model serving, prototype to deployment
  • Backend & real-time — Django, FastAPI, WebSocket video streaming, Postgres, MongoDB, Redis
  • Frontend — React, Next.js, TypeScript, Tailwind
  • Cloud & DevOps — Docker on NVIDIA GPU servers; GCP, Azure, AWS; GitHub Actions CI/CD
  • Systems — Linux, C/C++, Rust
Now Before
Ph.D. Research Assistant, University of Cincinnati · 2025–present Software Engineer → Web Team Lead, AiKYNETIX · 2023–2025 (promoted after 12 months)
Research Assistant (part-time), P&G Digital Accelerator · 2026–present Research Assistant, CV & ML, North American University · 2023–2025
Teaching Assistant, Python, ML & Applied AI, UC · Spring 2026 ACM club president — hackathons and workshops

Research areas

22 areas in 4 groups — how models learn, how to keep them secure and in bounds, what they are made of, and what happens once they act as agents.

Training & learning dynamics — How models learn, and what data, training, and fine-tuning do to their behavior.

RLHF FINE-TUNING LEARNING DYNAMICS TOKENIZATION

Safety, security & governance — Keeping trained behavior predictable, permitted, and auditable once it is deployed.

TRUSTWORTHY ML ALIGNMENT LLM SECURITY GUARDRAILS SAFETY EVALUATION PRIVACY & DATA GOVERNANCE FEDERATED LEARNING ROBUSTNESS

Models, modality & efficiency — What a model is made of, what it can perceive, and what it costs to run.

QUANTIZATION DIFFUSION VISION-LANGUAGE MODELS MULTIMODAL COMPUTER VISION

Agentic & retrieval systems — Models put to work: tools, retrieval, grounding, and serving.

AGENTIC AI REASONING MODELS RAG QUESTION ANSWERING MODEL DEPLOYMENT

Each area has context on mansurpro.netlify.app/research. Papers will be listed there as they become citable.

Selected work

AI, LLMs & agents
Project What it does
Multi-Agent Deep Research LangGraph supervisor/worker workflow: plans the question, gathers live evidence (Tavily), extracts structured findings with Pydantic schemas, scores source quality, and writes sourced reports with no-API fallbacks
voxide Offline voice control for developers in Rust — MiniLM sentence embeddings plus lexical matching map spoken commands to actions; 100% top-1 on a 40-utterance eval set
RAG Assistant Modular retrieval-augmented generation — ingestion, chunking, embeddings, ChromaDB semantic search, Flask conversational interface
DocuParse PDF/EPUB/MOBI → clean structured Markdown, with OCR, AI layout detection, and LaTeX equation handling
Data, cloud & full-stack
Project What it does
Retail Intelligence Platform Team project: end-to-end Azure analytics — customer lifetime value, churn risk, basket associations — FastAPI over Azure SQL + React dashboard with a SQL-backed cache
College Inquiry Chatbot Cloud-native conversational system on GCP that answers prospective-student questions
Prompt2Clip Cloud GPU pipeline for text-prompt-to-video-clip generation
Systems & fundamentals
Project What it does
DigitRecognizer MNIST classifier written from scratch in NumPy — forward pass, backprop, gradient descent, zero frameworks
DinoMind Evolution NEAT neuroevolution learns a timing-critical game with no gradients and no labeled data

All case studies at mansurpro.netlify.app/projects.

Tech I actually use

Languages & core  14
Python
Python
JavaScript
JavaScript
TypeScript
TypeScript
Java
Java
Rust
Rust
C
C
C++
C++
PHP
PHP
R
R
Bash
Bash
PowerShell
PowerShell
Regex
Regex
LaTeX
LaTeX
WebAssembly
WebAssembly
AI / ML / data  5
PyTorch
PyTorch
TensorFlow
TensorFlow
scikit-learn
scikit-learn
OpenCV
OpenCV
Anaconda
Anaconda
Backend & APIs  7
Django
Django
Flask
Flask
FastAPI
FastAPI
Node.js
Node.js
Rails
Rails
GraphQL
GraphQL
Nginx
Nginx
Frontend  14
React
React
Next.js
Next.js
Vue
Vue
Nuxt
Nuxt
Redux
Redux
jQuery
jQuery
Tailwind
Tailwind
Bootstrap
Bootstrap
Material UI
Material UI
Sass
Sass
Three.js
Three.js
SVG
SVG
Vite
Vite
Webpack
Webpack
Databases  7
PostgreSQL
PostgreSQL
MySQL
MySQL
MongoDB
MongoDB
Redis
Redis
SQLite
SQLite
Supabase
Supabase
Firebase
Firebase
Cloud, DevOps & OS  18
Docker
Docker
Kubernetes
Kubernetes
AWS
AWS
GCP
GCP
Azure
Azure
Cloudflare
Cloudflare
Heroku
Heroku
Netlify
Netlify
Vercel
Vercel
Linux
Linux
Ubuntu
Ubuntu
Windows
Windows
Git
Git
GitHub
GitHub
Actions
Actions
npm
npm
pnpm
pnpm
Yarn
Yarn
Tools & platforms  14
Figma
Figma
Photoshop
Photoshop
Notion
Notion
Postman
Postman
VS Code
VS Code
WebStorm
WebStorm
Vim
Vim
Replit
Replit
CodePen
CodePen
Stack Overflow
Stack Overflow
WordPress
WordPress
Gmail
Gmail
LinkedIn
LinkedIn
Bots
Bots

Contribution graph

Isometric 3D contribution graph for the last year

Rendered from the GitHub contributions API by assets/contrib.gen.py and refreshed daily by a scheduled Action — column height tracks that day's contribution count.

Stats

GitHub stats Contribution streak

Most used languages Most committed languages


Open to Research Scientist / Research Engineer roles and internships, ML & agent engineering roles, and collaborations.
[email protected] · LinkedIn · mansurpro.netlify.app

Hits

Pinned Loading

  1. Shareme Shareme Public

    Social media web app

    JavaScript 2

  2. Car-show Car-show Public

    Car show - rental cars

    TypeScript 1

  3. Prompt2Clip Prompt2Clip Public

    Prompt2Clip is a cloud-based application for generating high-quality videos from textual prompts using the Mochi video generation model. This project demonstrates the integration of advanced AI mod…

    Jupyter Notebook 2 1

  4. DinoMindEvolution DinoMindEvolution Public

    This project uses neural evolution to train a neural network to play the Google Chrome Dinosaur Game autonomously.

    Python

  5. machine-learning-projects machine-learning-projects Public

    Machine learning projects from Grokking Machine Learning by Luis Serrano

    Jupyter Notebook

  6. RAG-assistant RAG-assistant Public

    RAG-assistant is a type of AI assistant that uses Retrieval-Augmented Generation to provide more accurate and context-rich answers by combining a large language model (LLM) with an external knowled…

    Python