Standard Tariff: 100 EUR/h
Projects: 80 EUR/h
Phone: +420 775 775 559
LinkedIn: https://www.linkedin.com/in/jiri-bruijn/
Located in Prague, Czech Republic
Technologies: Python, Flask, FAISS, RAG, AsyncOpenAI, Torch, SentenceTransformer, Voice recognition, Javascript
Start: February 2025
Project duration: 3 months
After extensive research to find the best and most effective approach on how to train an expert medical model, I have build a custom LLM that was fed with academical articles in the field of myeloma / hematology. Three doctors, renowned hematologists, have been testing the LLM during the development based on which the LLM was tuned further. The training was finalized when all three doctors approved a set of benchmark questions. The model was incorporated in a custom made voice controlled chatbot of which the output is streamed to an Avatar API that live generates a video stream with lip-synced audio of the data the LLM returns.
The chatbot was presented at a congress in Paris on 18th May 2025 (https://comy.cme-congresses.com/). The chatbot was tested on a list of questions that three doctors were evaluated based on which the AI was tuned further.
Links:
- https://comy.cme-congresses.com/
- https://www.linkedin.com/posts/international-academy-for-clinical-hematology_aiach-ugcPost-7331274155109023745-kN4A
- https://www.linkedin.com/posts/controversies-in-multiple-myeloma-comy_meet-prof-richard-our-new-aiach-doctor-activity-7329543204926631937-FQO3
Technologies: Python, Flask, MS SQL, RAG, AsyncOpenAI, Javascript
Start: June 2024
Project duration: 9 months
A company produces pharmaceuticals on demand. It has an MS SQL database containing its recipes and what raw materials and half products are in stock and on which location. A certain product the company delivers can have a list of tens of raw materials needed, some are produced from other raw materials, which can be produced from other materials, etc. Some quantities are expressed in kilograms, some in millilitres, some in numbers and other units.
The company wanted to have an application containing a chatbot that the employees could use to know what is needed for the production of a certain product given a certain demanded quantity so that they would know what needs to be bought to fulfil a certain order.
I have created a system that understands if a user input is small talk, a database query, a calculation/custom function, or a follow-up question and based on this follows given flows in which operations are done, an LLM is asked, until in a final step an LLM is asked (again) to translate the results into human language (or in other cases predefined other formats). Results could be exported to CSV.
The chatbot was able to recognise over 20 different database query and custom function cases (what is in stock on a given location, how much is there of a certain product, a certain raw material, what do I need to fulfil an order, how much can I produce from what I have, list me all stocks, search for materials given a certain string, etc.).
Technologies / technical: Python, Tensorflow, Keras, Scikit-learn, Feature engineering, Custom loss functions, Creating, training and tuning of neural networka
Start: April 2018
Project duration: On going project (7+ years)
Numerai is a global artificial intelligence tournament to predict the stock market. Each data scientist downloads its data, builds models, and uploads his predictions. Because Numerai encrypts its datasets with structure-preserving encryption techniques, they are able to give away all of their data for free. This turns the stock market into a machine learning problem.
Since April 2018, I have been building successful and profitable models for Numer.ai.