BERLIN · AI ENGINEER · RESEARCHER
Khasim Bin Saleh
AI Engineer, Researcher, Agentic AI, Healthcare AI, Robotics
Can we trust what AI tells us?
I build AI systems for places where mistakes matter, from clinical software to robots, and I measure them before I trust them.
PROJECTS
Built, tested, measured
Finished work first, experiments after. Each page says what was measured and where the project stands.
Completed
Finished work.
03/2026 to 11/2026
Klindok
AI that never sends patient data anywhere.
Offline desktop software for private clinics. The doctor types bullet points, Klindok drafts the letter, checks every fact in the draft against what was typed, and exports a PDF. Nothing leaves the clinic computer.
100%
offline
5
kinds of fact checked
2
editions
- Python
- SQLCipher
- GGUF models
- PDF/A
- pytest

- Done
03/2023 to 08/2023
AI Tutor at Coschool
An AI tutor built from zero during my internship and shipped to 1,000+ beta users, with evaluation pipelines to keep its answers on topic.
1,000+
beta users
80%+
fewer off-topic answers
- RAG
- Milvus
- GPT
- Done
07/2025
Predictive Maintenance Pipeline
An end-to-end pipeline that predicts machine failures from sensor data. Five classifiers compared on F1 because failures are the rare class, with the full results table published, including the weak spots.
5
models compared
0.40
best F1
- scikit-learn
- XGBoost
- Docker
- pytest
- Done
04/2026
RAG Pipeline from Scratch
A retrieval-augmented generation pipeline rebuilt in plain Python, so every step between question and answer is visible. Evaluated with hit rate and mean reciprocal rank on a small labelled set.
88%
hit rate at k=3
0.79
MRR
- Python
- scikit-learn
- NumPy
- Docker
Ongoing experiments
Unfinished on purpose. No results claimed yet.
OngoingRole: AI features
Kiko
An interactive companion robot, built with a small student team. I work on the AI side. Early prototype.
Bench test. The face display, a camera and the compute board running together. The line on the screen says the vision models are not installed yet.
- Python
- Computer vision
- Robotics
RESEARCH
What I research, and why
CORE RESEARCH QUESTION
Can an LLM agent find the privacy risks in synthetic health data that standard tests rate as safe?
Hospitals and researchers share synthetic data in place of real patient records. Whether that is safe depends on the test used to check it, and popular tests can miss records that still point to a real person.
The same question runs through everything else I do: measure a system before you trust it.
Fig. 1 · Distance to the closest record
Each dot is a synthetic record. Each ring is a real person's record. A dot that lands on a ring can give that person away. Finding those cases is what my thesis is about.
- Real record
- Synthetic record
- Too close: 7 of 92
Move the pointer to audit a record. Illustrative data, not study results.
- In progress
MSc thesis
Agentic Privacy Auditing of Synthetic Health Data: Evaluating LLM Audit Agents against Attack-Based Re-identification Metrics
- Done
Enterprise AI agents
Compared four 2026 surveys where reported adoption ranged from 17% to 65%, and traced why.
- Done
Agent design patterns
ReAct, Reflection, Planning, Tool use and Multi-agent, matched to real deployments and to MCP and A2A.
Paper breakdowns
Published whenever I finish reading a paper properly. No fixed schedule.
- arXiv preprintMust read
The DCR Delusion: Measuring the Privacy Risk of Synthetic Data
The most popular privacy test for synthetic data passes datasets that membership inference attacks show are leaking.
JOURNEY
How I got here
Every step added one more test.
2020
Started B.Tech CSE
IoT specialisation at Lovely Professional University.
Education2023
ML intern at Coschool
Shipped an AI tutor to 1,000+ beta users.
Internship2025
Moved to Berlin
MSc Computer Science, Big Data and AI, at SRH Berlin.
MSc2026
Research assistant
Enterprise AI agents and agent design patterns. June to October.
Research2026
Built Klindok
Offline clinical documentation software with a fact check.
Done2026
Started the MSc thesis
Privacy audits of synthetic health data with LLM agents.
In progress2026
Kiko
A companion robot with a small student team. I work on the AI side.
Ongoing
ABOUT
From IoT tinkering to trustworthy AI
I started with a B.Tech in Computer Science at Lovely Professional University, with a specialisation in IoT.
At Coschool I built an AI tutor that reached 1,000+ beta users. That work taught me a model is only as good as how you test it.
I moved to Berlin for an MSc in Computer Science with a focus on Big Data and AI. Here I built Klindok, offline software that drafts clinical letters and checks every fact in them. Now I am writing my thesis on privacy audits of synthetic health data.
Alongside that I work on the AI side of Kiko, a companion robot, because I want to understand what changes when a model leaves the screen.
- Based in
- Berlin, Germany
- Studying
- MSc Computer Science, Big Data and AI, at SRH University of Applied Sciences Berlin. 2025 to 2027.
- Before
- Machine learning intern at Coschool, Hyderabad. B.Tech in Computer Science and Engineering, IoT specialisation.
- Languages
- Hindi (native), English (C1), German (B1, learning)
SKILLS
Technical depth
Honest levels: used in production, used in a project, or studying. A skill only gets a level when a project on this site backs it up.
- Production
- Project
- Studying
AI and ML
Models I train and compare when the data is messy.
scikit-learn
Predictive Maintenance
ProjectXGBoost
Predictive Maintenance
Project
Also worked with
- PyTorch
- TensorFlow
- Hugging Face Transformers
LLMs and Agents
Systems that answer, retrieve, and stay on topic.
RAG
Coschool AI tutor
ProductionGPT fine-tuning with guardrails
Coschool AI tutor
ProductionMilvus
Coschool AI tutor
Production
Also worked with
- LangChain
- OpenAI
- OpenRouter
- Gemini
- Azure Document Intelligence
- prompt engineering
- real-time speech (Deepgram STT, TTS)
Evaluation
How I decide whether a model is good enough.
Also worked with
- evaluation harnesses
- rubric-based LLM scoring
- evidence-based model and prompt selection
- statistical testing (Diebold Mariano, Friedman)
- time-series forecasting
- ensemble methods
Computer Vision and Robotics
Seeing the world, and learning to move in it.
ROS 2
Studying
StudyingRobotics
Studying
Studying
Also worked with
- OpenCV
- YOLO
- ArUco marker tracking
- real-time object detection
Engineering
The tools that turn a notebook into a system.
Docker
Predictive Maintenance
Projectpytest
Predictive Maintenance
Project
Also worked with
- Python
- SQL
- CUDA
- LaTeX
- Git/GitHub
- Linux (Ubuntu)
- Postgres
- Prisma
- FastAPI
- Streamlit
- React
- Power BI
CONTACT
Let's build AI people can trust.
Hiring, researching or building in healthcare AI or robotics? I would like to hear from you. LinkedIn is the fastest way to reach me.
- LLM Evaluation
- Healthcare AI
- Agents
- Robotics
Based in Berlin, Germany