About

Nightbox is a one-person biotech, incorporated in Wyoming, run out of the LA area. Started in February. The work is gene therapy for solid tumors, built around a chimera that fuses a human stress-ligand receptor with an elephant pseudogene effector.

Why

Cancer isn't really a disease in the way pneumonia is. It's a single cell forgetting that it's part of a body — an evolutionary reversion to the unicellular state our ancestors left behind. Most of the field treats this as a war against tumors. I think it's closer to debugging a long-running system that has corrupted its own cooperation rules.

Elephants are the case study that pulled me in. A hundred times more cells than us, and roughly a quarter the cancer mortality. The mechanism is partly TP53 copy-number expansion and partly a re-functionalized pseudogene called LIF6 that they use as a mitochondrial kill switch when DNA damage piles up. So the obvious question: can you write that kill switch into a single AAV9 construct and gate it on a tumor-restricted stress signal? That's the company.

How

One construct, 2,123 bp. NKG2D ECD up front (binds MICA/MICB on stressed and transformed cells, basically absent from healthy tissue). P2A linker. Codon-optimized elephant LIF6 in the back, mitochondrial localization signal preserved. Packaged in AAV9. No T-cell extraction, no autologous processing, no eight-figure manufacturing footprint. The IP whitespace on this specific fusion is genuinely clean — that's the part I checked first.

How the work actually gets done

It's me and an agent stack. That sounds like marketing — it isn't. The agent doesn't write the science. It compresses the operational floor: scanning patent corpora, cross-checking competitive pipelines, running freedom-to-operate against millions of documents, drafting regulatory paperwork, writing codon-optimization scripts. Three years ago each of those was a headcount. Now they're API calls.

The named pieces:

• Structural prediction: AlphaFold3, RoseTTAFold, ESM-IF for inverse folding
• Construct design: SnapGene + PyMOL for visualization, custom codon-optimization scripts
• Agent orchestration: Production Claude-based stack on Vercel Edge with Anthropic SDK, custom prompt-router, NFKC Unicode normalization, 22-pattern injection scanner
• Source repository: github.com/nightbox-llc/chimera-design-notes (methods + design documents, CC BY 4.0)

The agent isn't a co-founder. It doesn't get equity. It's plumbing — same way a wet-lab treats its automated liquid-handlers: named tools, version pins, methods open where it's safe to be open.

What I haven't proven yet

Most pre-clinical biotech sites duck this question. Here's the honest list of what could still kill the program and what the wet-lab is structured to find out:

• Folding behavior of elephant LIF6 in human cellular context — predicted stable, not yet expressed
• NKG2D ligand shedding by tumors as escape mechanism — known liability of NKG2D-targeted therapies
• On-target / off-tumor toxicity from NK and γδ T-cell engagement
• Optimal AAV serotype for solid tumor biodistribution — AAV9 is the working assumption, may not survive in vivo screen
• Whether the predicted ~99% TGI ceiling holds outside the PK-PD model in real syngeneic mice

Any of these can fail and force a redesign. That is the point of in vivo validation. The full pre-IND program is structured around these risks.

Founder story

Nightbox was founded by Artem Shakin, a 21-year-old computational biologist and technical founder based in the greater Los Angeles area. Self-taught in molecular biology and gene therapy, with an AI-augmented research workflow built over more than a year of daily iteration on chimeric receptor design, freedom-to-operate scanning, and patent strategy.

The thesis came from a simple observation. Elephants have ~100× more cells than humans yet die of cancer at a fraction of the rate — Peto's paradox, solved by nature through duplicated tumor-suppressor machinery (TP53 retrogenes, LIF6). Most of the field has been chasing autologous cell therapies with eight-figure manufacturing footprints. Nightbox asks the opposite question: what if you could write the elephant's cancer-resistance mechanism directly into a single AAV-deliverable construct, gated by a human stress-ligand recognition module so it only fires inside tumors?

The company is bootstrapped, sole-member, with no outside capital to date. Wyoming was chosen as the legal domicile for its mature LLC statute, charging-order protection, and zero state income tax on pass-through entities. Operations run from the LA biotech corridor for proximity to CROs, AAV manufacturing, and academic medical centers.

Artem is the sole employee, sole engineer, and sole decision-maker. The AI agent stack is research infrastructure — not staff, not co-founders. Wet-lab scientific lead recruitment opens Q2 2026.

Status

• ✓ Lead construct designed in silico (~99% TGI ceiling, PK-PD model)
• ✓ Freedom-to-operate scan complete; clean IP whitespace identified
• ✓ Provisional patent filing in progress
• → Construct synthesis + HEK293 in vitro expression validation — Q2–Q3 2026 (smoke test before in vivo)
• → AAV9 packaging at academic vector core or CDMO partner — Q3 2026
• → In vivo validation in 3 syngeneic murine models (CRC, melanoma, pancreatic) — Q3–Q4 2026
• → Pre-IND meeting with FDA CBER — H2 2026
• → IND-enabling toxicology (GLP, 6–12 months) — 2027
• → IND filing + Phase 1 first-in-human — 2027–2028 (subject to in vivo + tox completion)

Legal

Nightbox LLC · EIN 39-4373044 · Wyoming (incorporated) · Los Angeles area (operations), United States