JMH DataSciences

Joseph M. Hahn, Ph.D. · AI and ML consultant

Let AI do the tedious work,
your team has better things to do.

Retyping invoices into QuickBooks. Answering the same customer question the fourth time this week. Rebuilding the same report every Monday. Finding out an order slipped after the customer calls. While still waiting on the one person who can say which customers stopped ordering.

I build the AI that does that for you, typically via three forms:

  • AI agent that does the reading, checking, and drafting that your people are doing by hand.
  • AI assistant that closes the blind spots: ask your data a question in plain English, watch for the orders in danger of slipping, or generate a forecast you can actually hand someone.
  • Something custom built to move a number: win more quotes, ship more on time, buy better. The kind of edge nobody sells off the shelf, and is worth having before your competitor does.

My solutions run on your own data and are tested on your real examples before you trust them, and are priced in writing before I start. Your people then get that part of the week back to do the work they would rather be doing. And you get to see what you could not see before, while the number you care about starts to move.

Ph.D. physicist trained to take a hard problem apart and test whether the answer holds. Twelve years building production AI and machine learning, eight of them inside Oracle's AI Center of Excellence. Independent now: no vendor relationships, no referral fees. I deliver custom AI for small, mid-size, and enterprise businesses, bringing the judgment about what is worth building, and the working AI that then moves your numbers.

Joseph M. Hahn, Ph.D.

Which one is you?

You cannot afford to hire your way out of it

  • Orders and invoices still typed in by hand
  • The month-end report your team rebuilds from scratch
  • The req you have not filled since spring
See custom AI

Someone is selling you AI

  • A quote you have no way to price-check
  • Claims about your data that nobody has verified
  • The only AI experts in the room work for the vendor
See proposal review

Neither door quite fits? Tell me what is going on and I will reply within a business day. Headquartered in the Austin TX metro, with part of the year in the Knoxville TN area. Delivery is remote with on-site travel as needed.

Services

What I actually build.

Ten jobs I get asked for most, in three kinds. Each one is a specific piece of work, priced before it starts.

The work your people should not be doing.

The things you find out too late.

Something custom, built to move it.

Underneath, that is retrieval-augmented generation, natural-language database queries, single-agent multi-tool systems, and supervised machine learning, each picked for the job and benchmarked before it ships. See what each one looks like.

Engagements

Ways to work together.

One way in, and two directions from there.
Everything is priced in writing before it starts, and you can stop after any of it. No minimums, no long-term commitment.

Working Session

$250 · one hour

A paid hour on your actual problem, then a written page: what I would build, what it should cost, and what to do instead if building it is the wrong move. Credited in full against anything you book within thirty days.

The first conversation is free and always will be. This is the step after it, for when you want the answer in writing before committing to anything larger.

Book a working session

Someone is selling you AI

Independent judgment on work you are buying from somebody else. I am paid by you and by nobody selling to you: no vendor relationships, no referral fees, no reseller margin.

Most common

The Read

Read the proposal $2,500 · five business days

Send the proposal, SOW, or RFP response. You get back whether the architecture, the data assumptions, and the price hold up, what is missing, and what to ask before you sign.

Read the opportunity $12,000 · three weeks

Three weeks inside your firm. I find what data actually exists, rank where AI would move your numbers, and price each one to build and run. That cost line is what you hold a vendor's quote against.

Best for: an AI quote on the desk and nobody in-house whose job it is to judge it.

  • A written memo, plus a call to walk through it
  • Priced in writing before the work starts
  • Sometimes the answer is do not sign this,
    and if it is, I am not bidding on the replacement
Ask for a read How an independent review works

Owner's Technical Representative

$5,000 / month

Flat monthly fee. The cost is the same whether the pilot runs long or ends early.

Best for: a pilot where the only people who can judge the work are the ones being paid to deliver it.

  • Success criteria and acceptance tests agreed before work starts
  • Independent review at each checkpoint, and a call on whether to continue
  • I do not staff the build. Your team or the vendor implements, I hold the standard
Discuss pilot oversight What an owner's technical rep does

Your team is doing work a machine should do

I work directly on your systems. The retainer rate runs above the project rate on purpose: you are buying reserved capacity and continuity of context, not just hours.

Most common

One Build, Fixed Price

$5,000

One job, scoped in a short call and priced in writing before I start.

Best for: one concrete job, real data behind it, and no appetite for an hourly meter.

  • Delivered running on your own data, not in a slide deck
  • Tested on real examples from your business, with the score in writing
  • Handover session so your team can operate it without me
Scope a build What custom AI looks like

Ongoing Work

Project work $250 · per hour

Design, code, and documentation on your stack. Half a day or several weeks. Fixed-bid pricing once the scope is clear.

Retainer From $3,500 · per month

Ten hours at $350, or twenty at $300. Reserved capacity, a reply inside three business hours, and I hold working knowledge of your stack so we never restart from zero. Hours do not roll over, cancel anytime.

Best for: more than one job to do, or a team that wants a senior AI hand continuously engaged.

  • Direct work: design, code, and documentation
  • Standing weekly working session with your team
  • Final readout and handoff, so the work outlives the engagement
Discuss ongoing work

Custom engagements welcome. Multi-month projects, advisory boards,
and equity arrangements considered for the right fit.

Selected Work

Production AI for real businesses, not lab demos.

Recent enterprise use cases

AI-assisted use case demos

Recent Publications

How you know it works

Measured, not asserted.

Anywhere being wrong would cost you something, the work gets checked before you trust it, and you see the result. If the score is not good enough, you hear it from me rather than from a customer.

  • Your build is scored before you rely on it

    We agree a set of real examples from your business up front, run the system against them, and you get the result in writing before handover. Fifty past invoices. Fifty questions your team actually asks. Past months a forecast never saw.

  • The model is chosen by evidence, not by reputation

    Candidate models are judged blind on identical inputs by a stronger independent grader, so a cheaper one gets picked when it earns it. I published the method and the numbers.

    Eight LLMs graded blind

  • A result too small to mean anything gets called that

    I hold my own work to the same rule I would apply to yours, including the uncomfortable half: when a number is not yet strong enough to support a conclusion, the honest answer is to say so and keep measuring rather than to act on it.

About

A Ph.D. scientist who became an enterprise AI delivery expert.

I spent eight years at Oracle delivering AI engagements across manufacturing, oil and gas, finance, public sector, real estate, and retail. That work spanned customer discovery, technical design, and POC-to-production delivery. Before Oracle, four years delivering big-data ML systems on Hadoop for enterprise clients. And before that, two decades of peer-reviewed research in computational physics and planetary dynamics.

That combination is the differentiator: I am a scientist who can sit with the C-suite on Monday and write the production code on Tuesday. I work directly with technical and business stakeholders to turn ambiguous AI ambitions into systems that actually run.

As an independent practice, I work across industries and recommend the architecture that actually fits your problem. Vendor-neutral by design.

A small example of the method: this site was designed, built, and deployed with Claude Code in a matter of days. That same compression is what puts a working client prototype in weeks rather than quarters.

  • Ph.D. Physics, University of Notre Dame
  • 30+ peer-reviewed publications
  • 12+ years enterprise AI/ML delivery
  • 7 Oracle Cloud Infrastructure certifications

Capabilities

What I bring to an engagement.

The working stack I build with, and the judgment I bring when the build is someone else's.

Generative AI

  • Large Language Models (LLMs)
  • Retrieval Augmented Generation (RAG)
  • Natural-language-to-SQL (NL2SQL)
  • Multi-tool AI agents
  • Vector similarity search
  • LLM fine-tuning and GPU deployment
  • LLM evaluation and model right-sizing
  • Anthropic SDK / Claude API
  • AI coding assistants (Claude Code)

Machine Learning & Analytics

  • Python, SQL, pandas, NumPy
  • scikit-learn
  • Supervised ML on tabular data
  • Time-series forecasting and anomaly detection
  • Geospatial and spherical-geometry analysis
  • Graph algorithms

Data Platforms

  • Spark, Hadoop ecosystem
  • Jupyter notebooks
  • Git and version-controlled workflows

Independent Review

  • Proposal, SOW, and architecture review
  • Data-readiness assessment
  • Cost-to-build and cost-to-run estimation
  • Success criteria and acceptance tests
  • Checkpoint review and go/no-go calls

Consulting Practice

  • Customer discovery and use-case framing
  • Technical design and architecture
  • POC-to-production delivery
  • Executive stakeholder engagement
  • Technical writing and enablement content

Oracle Cloud Infrastructure (OCI)

  • 7 OCI certifications
  • Oracle Database
  • Oracle Autonomous AI Database
  • OCI Data Science
  • OCI Generative AI service
  • Oracle Analytics Cloud (comparable to Tableau and Power BI)
  • Conceptual fluency across major hyperscalers

Get in touch

Have an AI initiative that needs a steady hand?

Tell me what you are trying to build or untangle, and I will reply the same business day with a candid read on fit, scope, and next steps. No obligation.

Or write to me directly at joe.hahn@jmh-datasciences.com