Applied

PINNAL

Learned ranking platform for nucleic-acid therapeutics. Perceiver IO with KALAVAI-style cooperative LoRA fusion. Co-founded with Preethi Ravindranathan, clinical lead.

Why this exists

Nucleic-acid therapeutics are a fast-growing modality with thin tooling.

Antisense oligonucleotides, splice modulators, and gene-targeting candidates need better ranking infrastructure. Existing candidate ranking tools either oversimplify the biology through single-feature models or over-promise through black-box predictions with no clinical defensibility.

PINNAL is the platform we're building with both rigor and explainability - every ranking provably traceable to the input features that drove it.

Architecture
I

PINNAL-θ.

Perceiver IO cross-attention over heterogeneous biology - sequence, structure, tissue expression, off-target profiles. No tokenizer fit forced on biological diversity.

II

KALAVAI-style fusion.

Specialist LoRA adapters cooperate via token-level routing, drawn from Murai Labs' own KALAVAI research.

III

Schema to Consumer Audit.

Every input field provably affects ranking output. No silent placeholders. Built for clinical defensibility.

Benchmark Suite A
Co-founder

Preethi Ravindranathan - clinical lead. Molecular biology, peptidomimetics, androgen receptor research at UT Southwestern Medical Center, Department of Urology.

First-author publication in Nature Communications (2013): "Peptidomimetic targeting of critical androgen receptor-coregulator interactions in prostate cancer" - demonstrated novel D2 peptidomimetic with IC50 of 40 nM in mouse xenograft and ex vivo human prostate tumor cultures. Co-author in PNAS (2014) and Oncotarget (2015).

NVIDIA stack
Status
Related work

KALAVAI - Cooperative Specialist Fusion via Frozen-Layer MoE Routing.