Proof, not promises

Case studies that prove the technology

Real results across banking, government, healthcare, critical infrastructure and biometrics.

Case studies that prove the technology

Real results across banking, government, healthcare, critical infrastructure and biometrics.

Bank · Eastern Europe

33 → 43 tok/s

+0%

MultiCortex exceeded both the bank's performance target and the benchmark achieved by a Big Tech company.

Context

The bank operated a fleet of GPUs from different manufacturers and required an output speed of 39 tokens/s. Big Tech specialists worked for two months and reached 33 tokens/s.

MultiCortex approach

MultiCortex developed a special build using heterogeneous computing to optimize AI execution on the existing infrastructure.

Result

43 tokens/s, exceeding the 39 tokens/s target and the Big Tech result of 33 tokens/s — a 30% productivity increase.

Government project · Brazil

22 → 63 tok/s

Brazilian government AI project with significant performance improvement and lower computing cost.

Context

A major government AI project in Brazil, budgeted at approximately R$ 23 billion, initially operating at 22 tokens/s.

MultiCortex approach

MultiCortex applied its operating system and heterogeneous computing technology to optimize AI processing.

Result

63 tokens/s with MultiCortex OS. The corporate presentation reports a 65% reduction in computing cost, associated with lower environmental impact.

Healthcare

Private AI for healthcare

Specialized AI running on significantly more accessible infrastructure while producing more detailed responses.

Context

A test using the same prompt compared paid ChatGPT 5 on an estimated R$ 5 million machine with the MultiCortex solution running on a R$ 70 thousand machine.

MultiCortex approach

MultiCortex used its optimization technology to run a private AI solution specialized for healthcare.

Result

827 words versus 193 from ChatGPT 5, reaching responses up to 329% more detailed. All physicians consulted preferred the MultiCortex response.

Infrastructure · IBM Power 10

∅ → 15 tok/s

0 tok/s

AI inference enabled on IBM Power RISC architecture without a GPU.

Context

The Data Center partner was unable to run inference on IBM P10 machines, which do not have GPUs.

MultiCortex approach

MultiCortex created a special build for inference on IBM Power, using RISC architecture with MMA.

Result

15 tokens/s with inference enabled on IBM P10. The result was approved by an IT multinational with more than 15,000 employees operating in 13 countries.

Biometrics

600 → 30 analysts

Biometric engine optimization with significant impact on processing efficiency and anti-fraud operations.

Context

The client used Cognitec biometric technology and faced a false-positive margin of 0.001, with an anti-fraud desk of approximately 600 analysts.

MultiCortex approach

MultiCortex optimized the biometric engine processing using its AI and heterogeneous computing technology.

Result

Error margin reduced to 0.00000000001 and the anti-fraud desk reduced from 600 to 30 analysts. The corporate presentation reports a 99% increase in processing efficiency.

Hardware optimization

T4 · H100 · IBM P10

Optimization results across different architectures demonstrate the hardware independence of MultiCortex technology.

Context

AI environments using accelerators and processors from different manufacturers and architectures, including NVIDIA T4, NVIDIA H100 and IBM Power 10.

MultiCortex approach

MultiCortex applied hardware optimization and heterogeneous computing according to the characteristics of each environment.

Result

Recorded results include 33 → 43 tok/s on T4, 22 → 63 tok/s on H100 and inference enabled on IBM P10, reaching 15 tok/s without a GPU.

Your case could be next.

Talk to a specialist