BarzaX builds software that uses biological data to determine what to change to restore a function, make a molecule or improve a process, and to compute the consequences for the rest of the system.
For researchers and R&D teams working on interventions, molecules and bioprocesses, starting with enzymatic processes that are already measured. The long-term aim is to detect a decline early and design an intervention for one person, with the whole body taken into account.
Enzyme A is given a higher affinity for the shared cofactor and used in a smaller amount. Is it a good modification?
Both functions are above their minimum of 0.800 µmol/L/h.
The first engagement starts from an enzymatic process that is already measured.
The result needed, the measurements available and what can be changed, stated to Hermes, the assistant.
Biolographic supplies the data, sources, models and views.
Barzakh Engine computes the design and the decision.
The changes, their effects, their conditions and the next measurement.
Hermes calls Biolographic and Barzakh Engine through the Model Context Protocol (MCP).
Barzakh Engine is a library of more than 400 operations: kinetics, metabolic networks, resource allocation, reliability, experiment design. The same inputs give the same output, byte for byte. Some operations give exact values or guaranteed bounds.
Biolographic holds more than 3 million facts on human biochemistry, each with an identifier and a certainty: established, inferred, predicted, modelled or schematic. It also works on its own.
20 of 20
Yeast glycolysis. Replaying a published kinetic model of glucose-to-ethanol fermentation reproduces all 20 of its published steady-state values. A thermodynamic check finds 8 of its rate laws and equilibrium constants inconsistent with estimated reaction energies.
A published model, replayed: this checks the solver and the model, not new biology. Its time courses agree with two independent solvers to about 1 part in 10 million.
11 of 11
Known inherited metabolic disorders reproduced. Removing any of four urea-cycle genes stops urea synthesis from ammonia in the liver model; removing PAH stops tyrosine synthesis, as in phenylketonuria.
A test set of 22 known cases, written by us. The two cases where patients keep the function still pass. Nine fall on gaps of the model and are not judged; five of them were reclassified after a first run of 15 of 22. Model: human genome-scale metabolic model in four organ contexts.
133 of 140
Disease markers. Where a disorder's recorded marker sits on the blocked step, the rule gives the right direction (substrate up, product down) for 133 of 140 markers in 67 inherited disorders.
67 of 297 disorders with recorded markers can be tested this way. An independent checker replays every answer and rejects each proof with its sign or its step swapped.
Manuscript. The Law of Barzakh, screen edition 2 (2026). A function holds only while its work and the maintenance of its own machinery stay possible together on shared resources. The manuscript writes these demands as one budget, derives when it holds, and labels each result proved, established or predicted.
Patent application filed. Malaysia, PI2026007641: Measurement, design and control of enzyme-containing preparations.
Choose a preparation and its operating conditions.
Examine the intended effect and the functions it may compromise.
Integrate a person's measurements, follow their evolution, compare interventions.
Each project leaves models, data connections, control protocols and design methods that the next one reuses.
Bilal Medjani
“I'm tired of watching people around me get sick and die from diseases we should know how to prevent.”
Describe the result you need, the measurements you have and what you can change. No patient data.