Scientific computing, with the work shown.

Scienting checks and builds research computing for labs and companies in computational chemistry and biology, at posted, fixed prices. Platforms run the models. We check the results and build what they don't host.

See services and pricesSee how a check worksSee who does the work

studio.yaml
name: Scienting
staff: 1  # Alex M. Maldonado, PhD
help_from: colleagues, mentors
sells:
  - verification checks
  - computational builds
  - maintenance blocks
  - training days
prices: posted, fixed per project
code: Apache-2.0  # see Licenses
writing: CC BY 4.0

Services and prices

Posted, fixed prices

We take a few paid projects a year for labs, companies, and funders, each scheduled around a full teaching load. Every project starts with a free 30-minute call and gets a written, fixed quote before any work starts. If a project would go over its quote, we stop and ask first. Companies pay $150 an hour, and universities, nonprofits, and government labs pay $100. Maintenance is $75 an hour for everyone.

Prices at a glance

Posted prices for companies and for universities and nonprofits
Offer Companies Universities and nonprofits
First call, 30 minutes Free Free
Lite check of one result $1,200 $800
Verification check and report $1,500 to $4,500 $1,000 to $3,000
Build scoping, about five hours $750 $500
Build $6,000 to $18,000 $4,000 to $12,000
Maintenance, 10-hour block $750 $750
Fork assessment $1,500 $1,000
Rebuild of a forked tool $4,500 to $12,000 $3,000 to $8,000
Training materials, per hour $100 $75
Teaching, per day $2,000 $1,200
  • Your code, data, and unpublished results stay private. We publish nothing from a project without your written permission.
  • Your data stays on encrypted machines we control or with a secure cloud provider named in the contract, never on a public server, and is returned or deleted within 30 days of the project's end.
  • Code we write for you is yours. Any of our tools it uses keeps its own license.
  • We work only with software whose license allows paid work. Where your project depends on a commercial package, such as ORCA, Amber's pmemd, or AlphaFold 3, you supply the license and run those steps.
  • Checks include their compute. Builds list compute as a separate line, billed at cost with receipts.
  • For SBIR and STTR proposals, we can be a named consultant with a commitment letter, at the build or check price, payable if the award comes.

Verification checks

A verification check reruns your work and tells you what holds. Research code is often read by no one but its author, and small errors in it can change published results. A flipped pair of data columns in one lab's analysis program led to the retraction of five papers on protein structures (Miller, 2006). Trisovic and colleagues reran more than 9,000 R files from the Harvard Dataverse in a clean environment, and about 74% did not finish without an error; 56% still failed after automatic code cleaning (Trisovic et al., 2022). A second reader with time to rerun the work can catch problems like these before anything depends on them.

We check analyses before submission, data packages for partners and investors, and results produced on hosted platforms or by AI coding agents. Each report says what it covered and what it did not.

  1. Scope: you share the code, the data, and the results you expect, and we agree on what to check: a whole codebase, one pipeline, or the analysis behind a paper or report.
  2. Check: we rebuild the environment on a clean machine, rerun the work, and compare the output with your reported results. We review the methods and statistics, try the edge cases, and write a test for each thing we check.
  3. Report: you receive a written report that lists each problem, how to reproduce it, and how much it changes the results, along with the tests and the environment we used. It also says what we could not check.
  • If we cannot rerun the work at all, you pay only for the hours spent finding out why, up to the $1,200 lite check price.
  • You choose whether we fix the problems we find or your team does.

Expected costs for verification checks

These are starting prices. Companies pay the first price shown. Universities, schools, nonprofits, public research labs, and government agencies pay one-third less. Each project gets a written quote before any work starts, and any price change is posted here first.

Lite check
$1,200 flat$800 for universities and nonprofits
One figure or table rerun in a clean environment and compared with what you reported, in about 10 hours, with a short written report.
Check and report
$1,500 to $4,500$1,000 to $3,000 for universities and nonprofits
A fixed price, quoted once we have seen what you share. We expect most checks to need 10 to 30 hours. Any fixes you want from us are quoted separately.

Builds

In a build, we design the methods and the software for your question from scratch: docking and free-energy workflows, molecular simulation, quantum chemistry, cheminformatics and datasets for machine learning, machine-learned force fields, and genomics and transcriptomics pipelines, in Python, Rust, and Mojo. Builds run on your own data and machines when you need them to.

  1. Scope: we agree on the question, the methods, the data, and what you will receive, and write it up as a plan with milestones and a fixed quote. The plan is yours to keep, whether or not the project goes ahead.
  2. Build: each milestone ends with something you can run, with tests and a short note on what changed. We stop and ask before any milestone goes over its quote.
  3. Hand off: you receive the code, its documentation, an environment file that rebuilds it, and a report of the methods and their limits. Your team can run and change it without us.

Expected costs for builds

These are starting prices. Companies pay the first price shown. Universities, schools, nonprofits, public research labs, and government agencies pay one-third less. Each project gets a written quote before any work starts, and any price change is posted here first.

Scoping
$750 flat$500 for universities and nonprofits
About five hours: a written plan with the methods, the milestones, and a fixed quote for the build. The fee is credited in full to a build signed within 60 days.
Build
$6,000 to $18,000$4,000 to $12,000 for universities and nonprofits
A fixed price set by the plan, paid in milestones, with 30% to 50% due on signature. We expect most projects to need 40 to 120 hours.

Maintenance

Anyone who depends on a tool we built or maintain can buy maintenance in prepaid blocks: bug fixes, new Python and dependency versions, packaging, and help fitting the tool into your pipeline or cluster.

  1. Report: you send the bug or the question, with files that reproduce it. A report can stay private when it involves your data or code.
  2. Fix: we reproduce the problem, add a test that catches it, and fix it. The fix goes into the public release under the tool's own license, so every user gets it.
  3. Fit: block hours can also go to your setup, such as installing a tool on your cluster or reading your file formats. Larger integration work is quoted as a build.

Expected costs for maintenance

Every organization pays the same price for maintenance, because each fix goes into the public release that everyone uses.

Maintenance block
$750 for 10 hoursthe same for everyone
Prepaid at $75 an hour and used within a year, for any tool we built or maintain. We report what each block paid for.

Fork and maintain

Most research software was paid for once, by a grant, and maintained by whoever needed it next. A survey of omics software found that almost 28% of tools could no longer be reached at the address in their paper, and 28% of a tested sample could not be installed at all (Mangul et al., 2019). About a fifth of bioinformatics web tools never responded during a 133-day check (Kern et al., 2020). We fork tools like these and keep them working for as long as a sponsor funds the hours.

Each fork keeps the exact license of the original tool, and our changes are released under that same license. The original authors' copyright notices stay with their code, and anyone who could use the original can use the fork on the same terms.

  1. Assess: we check the license first, because code published without one cannot legally be copied or changed outside the hosting platform's own terms (choosealicense.com). Then we read the code, rerun its published examples, and write up what it would take to install, run, and trust the tool again. We contact the original developers before any fork goes public.
  2. Rebuild: we restructure the code and add tests that compare its output with the published results. Collberg and Proebsting could build the code behind only about a third of 402 computer systems papers within 30 minutes, and they asked funders to pay for this kind of repeatability work (Collberg and Proebsting, 2016). Each fix is offered back to the original project.
  3. Maintain: maintenance blocks go to fixing reported bugs and making the tool easier to use, through a Python API, a graphical interface, clearer logging, or better documentation. Each release reruns the comparison with the published results.
  • A fork credits the original authors and their paper in its README and in every release.
  • We work in the original repository instead whenever its developers prefer that.
  • The assessment report is yours to keep, whether or not the fork goes ahead.

Expected costs for fork and maintain

These are starting prices. Companies pay the first price shown. Universities, schools, nonprofits, public research labs, and government agencies pay one-third less. Each project gets a written quote before any work starts, and any price change is posted here first.

Assessment
$1,500 flat$1,000 for universities and nonprofits
About 10 hours: the license check, a build, one attempt to reproduce a published result, and a written report with a fixed quote for the rebuild. The fee is credited toward the rebuild if it goes ahead.
Rebuild
$4,500 to $12,000$3,000 to $8,000 for universities and nonprofits
A fixed price set by the assessment, paid in milestones. We expect most tools to need 30 to 80 hours, and we stop and ask before any milestone goes over its quote. Upkeep afterward uses maintenance blocks.

The Software Sustainability Institute publishes guides for people who maintain their own research software, and choosealicense.com explains how to pick a license that lets others keep the code alive.

Training

We design training materials to your standards and teach them to your team, in subjects drawn from Alex's research, teaching, and software:

  • genomics and transcriptomics, from sequence analysis and assembly to RNA-seq and differential expression
  • molecular simulation and enhanced sampling with OpenMM and GROMACS, using Amber and OpenFF force fields
  • quantum chemistry with Psi4, PySCF, and xTB
  • docking, cheminformatics, and computer-aided drug design
  • machine learning for molecules, from regression to machine-learned force fields
  • scientific Python, data engineering, and parallel and cluster computing, including high-performance code in Mojo
  • research software engineering: version control, testing, packaging, and documentation
  • checking computational results: rebuilding environments, rerunning analyses, and testing claims

You set the standard: the audience and its level, the format, the style guide, the accessibility requirements, and who reviews the drafts. We release the finished materials under CC BY-SA 4.0, so one client pays once and anyone teaching the subject can use and adapt them, with credit. Materials you want kept private are billed at our client rate instead. Your team does the work afterward and owns its results. We answer for the materials, not for what is built with them.

  1. Plan: we agree on the audience, the outcomes, the format, and your standards, and estimate the hours.
  2. Build: we write the lessons, exercises with automatic checks, slides, and example data. You review each draft against your standards.
  3. Teach: we run the sessions live, online or in person, as a workshop, an onboarding bootcamp for new lab members, or a course over several weeks. Self-paced materials can skip this step.
  4. Release: we publish the materials under CC BY-SA 4.0 after delivery and name you as the sponsor if you want the credit. Materials you paid to keep private stay yours alone.
  • Your own data and code never go into the public materials. Examples use public or synthetic data, and private sessions can work through your data.
  • Each set of materials is written new for the client who pays for it.

Expected costs for training

These are starting prices. Companies pay the first price shown, and universities, schools, nonprofits, public research labs, and government agencies pay the second. Each request gets a written estimate before any work starts, and any price change is posted here first.

Materials
$100 an hour$75 for universities and nonprofits
Planning, writing, and revising the lessons, exercises, and slides, against a written estimate in hours. Released under CC BY-SA 4.0 after delivery.
Private materials
$150 an hour$100 for universities and nonprofits
Materials you keep private instead of releasing them are billed at our client rate.
Teaching
$2,000 a day$1,200 for universities and nonprofits
A day of live sessions, online or in person, for up to 15 people. Travel, if any, is quoted separately.

Pro bono

We take one or two unbilled projects a year for noncommercial groups that could not otherwise pay, chosen so the results can be public. A case study says the work was unpaid, and the group is named only if it agrees.

Questions about services and sponsorship

hello@scient.ing

About

Alex M. Maldonado, PhD
Computational scientist

Alex M. Maldonado, PhD runs Scienting, a one-person studio that checks and builds research computing for labs and companies. He trained as a computational chemist and biologist and has written scientific software in academic labs since 2018, across molecular simulation, quantum chemistry, machine learning, cheminformatics, genomics, and transcriptomics, in Python, Rust, Mojo, and TypeScript.

During his PhD, he studied how to model chemistry in solution, including how solvation choices change predicted reaction mechanisms. He developed mbGDML, a framework for training size-transferable machine-learned force fields with limited quantum chemistry data. He then moved to drug design and structural biology as a postdoctoral researcher, and has maintained five tools developed and published by other researchers. The table lists each one with the people who wrote it.

He is a Teaching Assistant Professor of Biological Sciences at the University of Pittsburgh. Scienting is separate from that job. The University does not sponsor or endorse it.

Tools he has worked on

Tools Alex has worked on, with their authors and his role
Tool Developed and published by His role
mbGDML Maldonado, Poltavsky, Vassilev-Galindo, Tkatchenko, and Keith, Digital Discovery, 2023 Developer and first author
MolModa Kochnev, Ahmed, Maldonado, and Durrant, Nucleic Acids Research, 2024 Coauthor
Dimorphite-DL Ropp, Kaminsky, Yablonski, and Durrant, Journal of Cheminformatics, 2019 Maintainer, 2025; a creator of the 2.0.1 release
Gypsum-DL Ropp, Spiegel, Walker, Green, Morales, Milliken, Ringe, and Durrant, Journal of Cheminformatics, 2019 Maintainer, 2025
POVME Durrant, de Oliveira, and McCammon (1.0); Durrant, Votapka, Sørensen, and Amaro (2.0, 2014); Wagner and colleagues (3.0, 2017) Maintainer, 2024 to 2026
SubPEx Hellemann and Durrant, Journal of Chemical Theory and Computation, 2023 Maintainer, 2024 to 2025
WISP Van Wart, Durrant, Votapka, and Amaro, Journal of Chemical Theory and Computation, 2014 Maintainer, 2023 to 2025

When you use one of these tools, please cite its original paper and the version you used.

Tools

What we build in-house

We build these tools ourselves and publish their source. Each card gives the tool's license and status.

  • Programming+

    From an idea to Mojo code to the machine that runs it.

    A learning environment in three parts for Mojo, a Python-style language for high-performance code. The Bridge translates a procedure as you write it, from plain English through pseudocode to a specification and then labeled Mojo. In Developer, you write Mojo against the actual compiler and see each step and value. Hardware shows the registers, caches, and pipeline underneath. Nothing to install.

    Language
    TypeScript
    Form
    In the browser
    License
    Apache-2.0 code, CC BY 4.0 lessons
    Status
    Live at pplus.scient.ing

  • coursebank

    Course assessments as version-controlled data.

    Questions, exams, grading exports, and the statistics computed from them live as YAML and Parquet files in a git repository. coursebank assembles an exam, exports it for print or Canvas, reads the grading export back, and records how each question performed. The next exam starts from what the last one measured.

    Language
    Rust
    Form
    Command line
    License
    Apache-2.0
    Status
    In development

  • paperdesk

    A papers workbench that runs on your own machine.

    Find and fetch papers, convert each PDF into a structured package, and read it with linked citations. Tables copy out as CSV and equations as TeX. paperdesk holds no papers itself and keeps its own records beside your library, never inside it. Optional jobs that call Claude, Anthropic's AI model, use your own API key, show their price first, and run only when you confirm them.

    Language
    Python
    Form
    Web app
    License
    Apache-2.0 libraries, AGPL-3.0 app
    Status
    Working demo; new features paused

The vision

Why it works this way

Academic research software is often built for a specific project and rarely maintained after publication, leaving future users to handle the upkeep. In industry, expert time is costly and priorities follow company goals, so shared tools often go without support. Open-source software fills this gap and underpins tools worldwide, but many maintainers are volunteers, and reports describe persistent problems with unpaid work, burnout, and security. Low-quality reports and pull requests written with AI tools have added to the burden. The curl project ended its paid bug bounty in January 2026 to remove the incentive for low-quality security reports, "AI generated or not" (The Register, 2026).

Scienting builds tools meant to outlast any one grant or student project, and we build them for researchers and scientific teams who need long-term access. Every layer of our stack is designed to work with open-source tools, and commercial software can plug in to support adoption. We use open languages and well-documented file formats, so your work stays accessible even if a vendor changes its terms or its owner. Alex M. Maldonado, a full-time faculty member, builds and runs the studio. Client work and sponsorships pay for tool upkeep and, later, for paid contributors.

How it's funded

Who pays for what

Our tools are free to use. Labs and companies pay for the work around them: verification checks, builds, maintenance, and training, at the prices under Services. Sponsors can fund a named piece of work, such as a year of maintenance for one tool.

Tools

$0

Anyone can use our tools under each one's license. All but coursebank ship under open licenses that allow commercial use.

  • Read, run, change, and share our source.
  • Keep the license notices with every copy.

Services

Posted prices

Labs and companies pay for checks, builds, maintenance, and training. Universities, nonprofits, and government labs pay one-third less, except for maintenance.

  • Every organization of the same type pays the same price.
  • Fixes we make go into the public release.

We will publish a ledger of what comes in and where it goes, starting with the first payment.

AI in our work

How we use Claude

We use one family of AI models: Claude, a set of large language models made by Anthropic, mostly through its coding tool, Claude Code. Claude drafts code, tests, and documentation under Alex's direction. Alex designs each tool and each project, decides what Claude is asked to do, and reviews and tests every change before it is released or delivered.

  • Accounts: our own tools are built under a Claude Max subscription, governed by Anthropic's Consumer Terms of Service, with model training on our sessions turned off. Work on a client's code or data runs through a separate Scienting account on Anthropic's API, governed by Anthropic's Commercial Terms of Service, which state that Anthropic "may not train models on Customer Content." Client code and data never go into a tool that trains on them.
  • Authorship: US copyright protects only work by a human author (Thaler v. Perlmutter, D.C. Cir. 2025; certiorari denied 2026), and the US Copyright Office protects a person's selection, arrangement, and modification of AI-assisted material (Copyright Office report, 2025). Alex's design, direction, review, and revision are that human authorship, and our commit history records them. Any copyright registration we file will disclose AI-generated material that is more than de minimis, as the Copyright Office requires.
  • Responsibility: Scienting is responsible for work made with Claude to the same extent as for any other work it delivers, under the terms of each contract. Each contract states that AI tools may be used and names them.
  • Projects without AI: a client can ask that no AI tool be used at any step of a project, from planning to delivery, and the contract will say so. The work is billed at the same rates, and the quote includes the extra hours it takes to write, test, and document everything by hand. The starting prices in Services and prices assume we work with Claude.
  • Contributors: our contributor license agreement will require outside contributors to disclose any AI use and to confirm that they reviewed, and can answer for, everything they submit.

Licenses

What you can do with our work

Each kind of work has its own license. Apart from coursebank, our code ships under licenses approved by the Open Source Initiative, so it is open source. Our writing uses Creative Commons licenses.

Our licenses by kind of work
Work License What it allows
Libraries and tools Apache-2.0 Use, change, and share for any purpose, including commercial use, with the license and notices kept. It includes a patent license.
Packaged apps AGPL-3.0-or-later The same freedoms, and anyone who runs a changed copy for other people over a network must offer them its source.
Writing, documentation, and reports CC BY 4.0 Share and adapt for any purpose, with credit.
Course and training materials CC BY-SA 4.0 Share and adapt for any purpose, with credit, keeping adaptations under the same license.
Data CC0 1.0 or CC BY 4.0 CC0 where a repository requires it, with no conditions. CC BY elsewhere, with credit.
Forks of other people's tools The original license Our changes use the same license as the original code.
Work for a client The contract Code we write for you is yours.
  • Code and writing apart: Creative Commons recommends against using its licenses for software, and they do not license patents. We keep code and writing in separate files and label each one.
  • Released versions: a license applies to the version it shipped with.
  • Contributions: outside contributors will sign a contributor license agreement before their first change is merged, so every file keeps a clear owner.