SAS Hackathon 2026 · Manufacturing

Lantern

SAS Viya works out what needs doing in a factory each week. Lantern is where the team does it: every action goes to the person who owns it, they talk it through, do the work and record what they did, and SAS checks whether it worked.

Try the open demoHow it works

The demo opens in a new tab as someone at Meridian Components, a made-up manufacturer. Nothing you do there is saved.

Lantern's Actions page: a list of reorders ranked by money at risk, and one open action with its owner, due date, progress and a suggested message to the supplier
The planner's inbox. SAS ranks the week's actions by money at risk; each one has an owner, a reason and a due date, and moves from raised to taken on, resolved and checked by SAS.
Department files in
19
Cleaning rules
40
Rows they fixed
28,098
Actions from one weekly run
306
Languages
6

From the build's own output, on the made-up company's data.

Why I built it

For my father

My father owns a factory in the south of Vietnam. It is his life's work. When I came home after years in the U.S., I went to see it again, and I wanted a way to talk with him about it. I had been away a long time, and the business was something we could share.

He is a traditionalist. He believes in hard work, in being thorough, and in what people put in. So do I. But technology keeps moving: in machines, in software, in how people work and how they read their data. Much of his company still runs by hand, and on the memory of people who are close to retiring.

So I thought about my own work. At the trucking company where I started, we built one place that showed everyone the bottleneck costing the business the most, so the right people could act on it early. It taught me that a number only matters once someone owns what happens next. That became Lantern.

Funny enough, my first SAS tool was JMP, back when I was learning statistics in college. I never thought it would lead here.

Why Lantern

The name comes from Hội An, an old town in central Vietnam, where the river carries lantern light past houses older than any of us. A lantern does not chase the dark away. It just shows you the next step, and lets you walk it yourself.

That is what I want Lantern to do. SAS lights the next step; the person who owns it takes it.

So this is a love letter to my country. I hope we keep our light: guiding others gently, keeping the beauty that is ours, and still holding our own against the world.

Tea

The problem

Workbooks everywhere

A mid-size manufacturer runs on department spreadsheets. Planning, purchasing, the plants and finance each keep their own numbers, the reports disagree, and a loss is usually found after the money has gone.

Analytics can say what is about to go wrong. Then a dashboard shows it to everyone, and nobody in particular does anything.

The idea

One owner each

Let SAS clean the files, forecast, score the risk and decide. Then hand every decision to one person, in a workspace where people, SAS and an assistant work on it together.

And close the loop: what people did goes back to SAS, which measures what it was worth.

How it works

The action desk

A toy version of the weekly loop. Press the buttons and watch where each action goes.

Files cleaned
0
Actions raised
0
Resolved
0
Nobody acted
0
Checked by SAS
0

  1. 01

    Files

    Nineteen department spreadsheets, each with its own habits: money typed as text, mixed date formats, local shift codes.

  2. 02

    Clean

    SAS flows apply 40 logged rules, so every fixed cell can be traced back to the file it came from.

  3. 03

    Decide

    Models forecast demand and score stockout risk. A published decision turns each risk into an action with a reason and money at risk.

  4. 04

    Assign

    Each action has one owner. A shift lead sees their own plant and shift; the chief executive sees everything.

  5. 05

    Act

    The owner takes it on, talks it through with the team, does the work, then records what they did.

  6. 06

    Check

    Every decision goes back to SAS. Once an action's window passes, SAS checks whether the problem was avoided.

Along the way

Notes from the build

A few weeks, mostly at night. What it was like, in the order it happened.

  1. In Viya

    A calm place to work

    The first thing I noticed about SAS Viya was how calm it is. Clean screens, one job at a time, nothing shouting for attention. I expected enterprise software to feel heavy. Viya felt smooth, almost elegant, and I liked being in it.

  2. In Viya

    Nineteen files that disagree

    Meridian arrived the way real companies do: nineteen department files that do not agree with each other. I cleaned them in SAS Studio flows, which read like a drawing of the work. Forty logged rules fixed 28,098 rows, and every fix was written down.

  3. In Viya

    The code was nice too

    I wrote more SAS code than I expected, and I enjoyed it. The program, its log and its results are one click apart, so you always know what just happened and why.

  4. In Viya

    A small lesson

    Data Explorer guessed column types from the first twenty rows and quietly blanked 204 values. Nothing failed; the numbers were just wrong. Now I count rows and empty values after every load.

  5. In Viya

    Where it clicked

    Model Studio lined the pipelines up side by side, and a forest won with a held-out AUC of 0.939. Then Intelligent Decisioning turned models and rules into actions with an owner, a priority and a deadline. That was the moment it stopped being a chart and became somebody's job.

  6. 20 Oct

    The workspace

    With SAS solid underneath, Lantern came together: actions, reports, data and Ask, in six languages. I made Ask check every number against the SAS table it came from. If it cannot, it says so.

  7. 23 Oct

    Closer to home

    Until now Lantern ran on a small computer on my desk, through a tunnel. I moved it to servers in Singapore, and from Vietnam it suddenly felt instant.

  8. 29 Oct

    Packing up the lanterns

    Our Viya environment would be deleted after the 31st, so I exported everything: 75 tables, 180,189 rows, the programs, the reports and the champion model.

  9. 30 Oct

    Leaving the door open

    An open demo, so anyone can look around as someone at Meridian. What they change stays in their own browser.

  10. 31 Oct

    The lights stay on

    Viya closed for us today. Lantern still runs on its last results, and I am a little proud of that.

Screens

Inside Lantern

What the planner sees in the open demo, in week 8 of a 13-week replay.

Lantern's Today page: a one-line weekly brief, actions by priority, and a band showing where money is stuck between inbound, stock, plants and margin
Today opens with one sentence: how many actions SAS raised this week and what is at risk if nobody acts.
A map of supplier lanes from Japan, South Korea, Taiwan, China, Thailand, India and Malaysia into Vietnam, beside a map of four plants sized by money at risk
Further down, the same week as a map: each supplier lane coloured by how often it arrives late, each plant sized by money at risk.
Ask Lantern answering 'Which suppliers should we call this week?' with a list of suppliers and amounts at risk, the SAS table it used, and a note that every figure was checked
Ask answers only from SAS tables the person is allowed to read, names the table, and checks every figure against it before it is shown.
Lantern's Analytics page: 13 billion dong at stake a year and 4.9 billion recoverable, with a bar chart by area
The Visual Analytics story, redrawn by Lantern from the same SAS tables for everyone without a Viya seat. The table behind each chart is named under it.
Lantern's Results page: value protected, problems avoided, happened anyway and nobody acted, by team and by week
Results is SAS's side of the loop: after each action's window passes, did the problem happen anyway? Nobody acting is counted too.
Lantern's Data page: every dataset with its tier, row count, which roles can read it and whether the viewer can
Every table Lantern shows, where it comes from, and who can read it. The planner can read stockout risk, but not sales and margin.

Screens

Inside SAS Viya

The other half. These were taken in the hackathon's Viya environment before it closed on 31 October 2026.

A SAS Visual Analytics report, Meridian Control Tower, with gross profit by product family and monthly revenue and gross profit
Meridian Control Tower in SAS Visual Analytics, with tabs in English and Vietnamese.
A SAS Model Studio pipeline with imputation, variable selection and seven models feeding a model comparison
The stockout risk pipeline in Model Studio: seven models trained on the same data, then compared.
Model Studio's comparison table with the Forest model starred as champion
Forest won, with an area under the curve of 0.94 on held-back weeks.
A rule set in SAS Intelligent Decisioning that sets the action type to reorder when stock is below the reorder point or the model says a stockout is likely
The rule set that chooses an action. Every rule writes a reason code, which is the reason a person reads in Lantern.
A decision flow in SAS Intelligent Decisioning: start, the Forest model, two rule sets, end
The published decision: the champion model scores risk, then two rule sets pick the action, its priority and its owner.

Design

How it looks

Lantern's home page: the headline 'Know what needs doing. Get it done, together.' beside a dotted map of Asia with supply routes into Vietnam and three example actions
The front page. Supply lanes by air, sea and road run into four Vietnamese plants, and each one ends in an action with a person's name on it.
  • A calm product, not a wall of charts. Each page leads with one sentence that says what matters this week.
  • Light and dark, soft colours, no shadows. Money in đồng and place names in Vietnamese, because that is where the plants are.
  • Under every chart, the SAS table it came from. A number nobody can trace is a number nobody trusts.
  • Six languages from one switch: English, Vietnamese, Korean, Japanese, and Simplified and Traditional Chinese.

Why it is built this way

  • SAS decides, and nothing around it recalculates. Lantern shows SAS's numbers and never makes its own.
  • Viya seats go to the few people who build the analytics. Everyone else acts in a plain web app.
  • Who sees what is enforced by the database itself, row by row. Decisions are append-only, so the record cannot be edited later.
  • Ask can only answer from tables the person may read, and every figure is checked against them before it appears.
  • When the hackathon's Viya closed, the reports were redrawn from the same tables, so the product kept working.
  • The open demo reads a saved copy and keeps a guest's changes in their browser. It cannot reach the database or a language model.

How it feels

This one was a sprint: a whole product, a SAS build behind it and a film for the jury, in a few weeks with my team.

My favourite part is the loop. SAS decides, people act, and SAS checks whether it worked. Most dashboards stop at the first step.

Tea

What it runs on

SAS Viya
Studio flows, Model Studio, Model Manager, Intelligent Decisioning, Optimization, Visual Analytics, a weekly job
App
Next.js on Vercel
Data
Postgres with PostgREST and row-level security
Email
n8n and Resend
Ask
A language model, answering only from SAS tables

Open demo

Try it yourself

Pick someone at Meridian and use Lantern as they do. Take an action on, resolve it, leave a comment, ask a question.

Try the open demo
Lantern's sign-in page with an open demo card listing four people at Meridian Components to view the product as
Pick a person, no account needed. What you change stays in your browser and is never saved.

All workNext: The decoding line