Press & Media

Press kit and media resources.

Founder bio, technical claims summary, and downloadable assets for journalists, researchers, and analysts covering Bhala’s work on programmable embeddings.

press@mail.bhala.aiResearch overview

Technical claims

What Bhala has built, stated precisely.

Each claim below is documented on the benchmarks page with methodology, dataset references, and reproducibility notes. Citations appreciated.

An encoder pretrained on a single language transfers across 17

The Bhala encoder is pretrained on isiZulu alone. On MASSIVE 60-way intent classification, using the field's strictest representation-quality test, it reaches 73.2% on Swahili (above GPT-4o zero-shot at 70.6%), 72.5% on Korean, 69.7% on Hindi, and 66.5% on Amharic — none of these languages were in the training corpus. No labels, no fine-tuning: only a small monolingual sample is used to adapt the input per language. Results clear 38–43× over random across all four languages. To our knowledge, no other published model satisfies this exact test condition (zero target-language training) on typologically distant languages. Source: Mhlambi 2026, “Structure is all you need.”

Operator transfer at 95–100% on held-out test data

A sentiment direction estimated from contrastive pairs in one language applies zero-shot to others without re-estimation. Tested pairs to date: Zulu→Swahili (100%), Zulu→Xhosa (95%). Intent operators (book→cancel, alarm→calendar) transfer at 100% on tested pairs. Operators compose, invert, and produce a signed audit record per application. Cross-family transfer of the same operator class is in progress.

Architectural inductive bias accounts for ~80 percentage points of accuracy

A standard pipeline approaches random performance on this task. Bhala's architecture closes that gap to production-grade accuracy. The contribution of each architectural component is documented in the technical paper, available on request.

100% strict-flip on 28 protected dimensions across canonical fairness benchmarks

BBQ, StereoSet, CrowS-Pairs, and WinoBias. 15,966 sentence pairs covering age, disability, gender, nationality, physical appearance, race, religion, sexual orientation, socioeconomic status, and more. An independently-trained classifier accepts every shifted embedding as belonging to the anti-stereotype class. Methodology and per-benchmark detail published on the benchmarks page.

Hate-speech production: 11 corpora, HateCheck 0.90, TweetEval-hate 0.77

Bhala is evaluated jointly across 11 hate-speech corpora — HateCheck, CONAN, Civil Comments, Berkeley MHS, SBIC, DynaHate, TweetEval-hate, TweetEval-offensive, HateXplain, Stormfront, Hate-Speech-18 — for ~134K labeled examples. HateCheck AUROC 0.90 matches or beats HateBERT (0.85–0.88) and HateXplain BERT (0.83), which were fully fine-tuned on hate data. TweetEval-hate AUROC 0.77 — Twitter performance without any Twitter pretraining. Live in production since 2026-05-02.

CPU-deployable: <50ms inference, no GPU

Bhala runs on consumer CPU with sub-50ms single-query latency. No GPU required for inference. Designed for edge deployment in low-resource settings, but the same model serves the hosted API.

Founder

Sabelo Mhlambi

Founder, Bhala AI.

Sabelo Mhlambi is the founder of Bhala AI. His background spans NLP engineering and AI ethics: software engineering at Mapbox and Iris.tv, and fellowships at the Berkman Klein Center for Internet & Society (Harvard), the Carr Center for Human Rights Policy (Harvard Kennedy School), Stanford PACS, and TechCongress. He is the founder of Bantucracy, a research initiative on Ubuntu ethics and AI.

Bhala is the commercial vehicle for a decade of work at the intersection of low-resource language technology, decolonial AI, and representation theory. The company's central technical claim — that a sensitive concept's influence can be located inside a model's learned representations, removed with minimal collateral damage, and proven removed with a record an auditor can check, on models Bhala did not train — emerged from that lineage.

Assets

Logos and photos.

Right-click to save, or contact press@mail.bhala.ai for additional formats.

Founder photo (square, 1024×1024)

PNG

Founder photo (square, 512×512)

PNG

Bhala wordmark (banner, 3000×1000)

PNG

Bhala icon

SVG

Paper & reproducibility

Methodology, code, weights.

Preprint forthcoming

Structure is all you need

Mhlambi · April 2026 · Bhala AI

Bhala’s architecture and training objective, with component-level ablations and cross-family transfer results across 17 MASSIVE target languages. Argues that linguistic inductive bias carries information currently paid for in scale, and that the parameter count required for a given capability is a function of the task, not of a scaling law. Preprint link added here on release.

Benchmarks page

Per-benchmark numbers, evaluation protocol, dataset references, and links to the audit harness.

View benchmarks

Live demo

Apply sentiment, intent, and bias-removal operators to your own text. See the audit receipt for each call.

Open demo

Press inquiries

For interviews, technical briefings, or independent reproduction access, write to press@mail.bhala.ai. We respond within two business days.