doradus-research

Formally-verified arXiv paper

You provide the question, the data, or a draft. We run the investigation through our autoresearch pipeline, formally verify the mathematical claims via Lean 4, structure the paper for peer-review survival, and deliver compile-clean LaTeX source plus all figures, supplementary material, and reproducibility artifacts. Submitted to arXiv under your authorship.

Patent services

Prior-art search across patent databases and the scientific literature, novelty + non-obviousness assessment, claim-drafting support, and freedom-to-operate analysis. Research-grade, not legal advice — designed to walk into your patent attorney's office with the technical work already done.

Custom LLM training & fine-tuning

LoRA / QLoRA adapters, full fine-tunes, continued pre-training, RLHF / DPO / GRPO post-training — on your data, on our cluster. Domain-specialized models for scientific, medical, financial, or proprietary-corpus work. Deliverable is the trained weights (HF format) plus reproducible training config and an evaluation report.

3D scanning, reconstruction & printing

Photogrammetric + structured-light capture, deep-learning 3D reconstruction, mesh-to-CAD conversion. In-house printing on a Modix large-format printer (4 ft × 2 ft × 2 ft / 1219 × 610 × 610 mm build volume) for oversized parts, and a Bambu Lab multicolor printer with AMS for fine-detail multi-material work. Object in, printed part + digital assets out.

Paid inference coming soon

Membership-tier access to inference on our model fleet — frontier MoE rotation pools, scientific-domain specialists, on-prem privacy. More info shortly. In the meantime, perfct.ai is our billing surface.

Both deliverables span the same domain depth — the autoresearch pipeline auto-routes claims to specialized models (chemistry, materials, biology, finance) for verification:

Chemistry & molecular

SMILES + IUPAC claim verification, reaction prediction, computational-chemistry method critique (DFT, MD, ab initio), stoichiometric and formula-level cross-checking against published literature.

Materials & crystallography

Lattice constants, space-group analysis, unit-cell geometry, X-ray diffraction interpretation, CIF parsing, materials-property prediction and verification against the published structural record.

Physics

Condensed-matter, statistical mechanics, computational physics — DFT / MD / Monte Carlo method critique, simulation-result review, dimensional analysis, cross-checking measured-vs-modeled values against the published record.

Biology, genetics & proteins

Protein-sequence + structural analysis (folds, helices, beta-sheets, quaternary topology), gene-level expression and variant studies, peptide / fusion-construct verification.

Medical literature review

Claim-level verification across clinical and translational literature, cross-source agreement checks, mechanism-of-action validation. Not clinical advice — investigator-grade source synthesis for systematic-review and meta-analysis support.

Finance × science crossover

Biotech and pharma due-diligence, drug-pipeline review, materials-finance, energy-transition technical risk. Same scientific-verification depth, layered onto financial peer review.

Bespoke deep dives

Your domain is probably also weird enough to investigate. If your question doesn't fit the categories above, ask — odds are we'll find it interesting too.

  1. Submit the scope You describe the question, the desired deliverable (arXiv paper or patent package), what you've already done, and your timeline.
  2. Feasibility + quote within 48 hours We respond with our read on the question, what we'd investigate, the deliverable scope, and a fixed-price quote. No surprise hourly bills.
  3. Investigation + formal verification Autoresearch pipeline runs the literature + claim sweep; mathematical claims get encoded and machine-checked in Lean 4; figures + tables generated from reproducible source. Intermediate findings shared so you can steer.
  4. Deliverable + debrief For papers: compile-clean LaTeX + figures + supplementary + Lean proofs + arXiv submission under your authorship. For patents: prior-art memo + novelty/non-obviousness assessment + claim drafts ready for your patent attorney. Plus a 60-minute debrief call.

Investigations run through our internal autoresearch pipeline — multi-agent retrieval, cross-source verification, scientific-domain augmentation — then get walked through by an operator who knows the cluster, the bugs, and the upstream codebases firsthand. The deliverable is a written report with reproducible evidence and concrete recommendations. Not a transcript, not a wall of agent traces.

No public sample reports yet — this is a new offering, and customer engagements stay private by default. For a sense of the depth and voice we bring to technical investigations, see the operator blog. Those posts aren't paid-research output; they're the same operators writing about the same cluster on their own time.

Ready to commission an investigation

Checkout, account setup, and engagement intake all happen on perfct.ai — our billing surface.

Start a research engagement No prepayment until scope + quote are agreed.