JasonCEC

My company[0] has been considering a verification service to check on origin and sourcing claims using flavor signatures in "craft" products.

Currently, all of our tools are used by beverage producers (beer, coffee, spirits) internally for quality and process optimization - it would be great to apply our models and create a subscription reporting and claim verification.

[0] www.Gastrograph.com

Edit: clarity and repetition.

Also, Any ideas? What would you like to see in a service like this?

No... we're a full fledged product company; the basis of the system is our core set of algorithms that work across 99% of beverages, which we've turned into a set of applications.

The implementation work is part of our Professional Services, which is standard in Enterprise contracts - that could be viewed as consulting, but its targeted for getting our product into and across the large enterprise.

Hey Peter,

Thanks!

Radio Carbon dating is used for authenticating rare vintages of wine[0], but it's a minor test for the most part.

In modern wine collecting, almost all of the fakes are focused around a few _Gran Cru_ lots of small vineyards, like chateau lafite or Petrus.

It's actually not that hard to find wine from the 60's through 70's (or earlier. Counterfeiters use real lesser wine from the right time period, and forge the label and bottle - assuming they can't find an empty bottle from the actual production run! These wines easily sells for upwards of ~$10,000 to 3 - 4 x that.

So radio carbon dating is only marginally useful in wine; you need to use stable isotope ecology to see if the vintage matches the right terroir[1].

To make it even harder, most of the top vintages of French wine illegally imported grapes from neighboring regions to increase their complexity and overall flavor - this was a known but secret practice for about ~100 years, and since the French AOC[2] cracked down on it, there are concerns that modern wine lacking this complexity from blending will not age as well over ~20+ years.

I hope this information is helpful!

[0] http://www.acs.org/content/acs/en/pressroom/newsreleases/201...

[1] https://en.wikipedia.org/wiki/Terroir

[2] https://en.wikipedia.org/wiki/Appellation_d'origine_contr%C3...

On the off-hand chance that anyone here is super obsessed with tea, I'm the founder and director emeritus of The Tea Institute at Penn State[1], and blog here: [2].

I now run a flavor profiling and quality intelligence company based off my research in the Institute.

[1] http://sites.psu.edu/teainstitute/

[2] http://cultofquality.com/

Agreed - State College internet is dismally slow and unreliable for a town of ~60,000 college students, grads, and employees.

Thanks for the reply!

We work across the industry in alcohol production, so small craft to giant global sized companies.

A lot of what we do is implement better data acquisition from sensory, production, distribution, and demographics during our implementation phase, before our prediction or optimization models have a good enough baseline to run - we are definitely not a sane software shop!

ERPs are a threat but not direct competition to us - we're hardcore "best of breed" software, and it will be interesting to see how those large companies respond to us in the future...

The short answer is: _you don't_. That's not how machine learning works at scale.

To quote from a great recent article: (I can't find the link, sorry)

It's one thing to create an excellent fraud detection model in R, and quite another to build:

  - Fault-tolerant ingest of live data at scale that could represent fraudulent actions
  - Real-time computation of features based on the data stream
  - Serialization, versioning and management of a fraud detection model
  - Real-time prediction of fraud based on computed features at scale
  - Learning over all historical data
  - Incremental update of the production model in near-real-time
  - Monitoring, testing, productionization of all of the above

You don't build a data team out of a single person and tack on an easy model to build a company - it takes a team to build a real data project, and those teams are hard to find, hard to recruit, and hard to make successful.

Evan beat me to it :)

Evan is our Lead Engineer, and I'm the CEO.

We believe vertical integrated SaaS focused on prediction will reshape a lot of industries.

We've been using Phabricator[1], a spin-out of Facebook, and love it. It's perfect for technical teams, and usable enough for the rest of the company.

[1] http://phabricator.org/

Couldn't agree more! We're in a niche industry (food and beverage _production_) as a "vertical-specific SaaS" and there are 1) few players 2) immense demand for software that can do anything and everything.

My takeaways from operating in a "vertical-specific SaaS" are:

1) Lots of value to be captured by making something for specific vertical and then growing within that vertical

2) Owning the data end-to-end is our "network effect" or "secret-sauce" or whatever VC Buzzword they ask about (but is truly valuable)

3) Domain knowledge is critical. We need to know _a lot_ about the beverage product being produced and speak fluently about it to have any credibility and any chance of addressing their pain-point.

Just as native as French Wine vs Italian wine. Ingredients and production process are what create the product - the optimization is local to the potential flavors inherent from those ingredient processed in that way.

My company[0] works in the alcoholic beverage industry, and we've been watching the market for Asian and other native alcohols evolve.

I strongly believe that, more than sake and sochu (not soju), Chinese native spirits have a chance to take off globally, even if only because of demographic momentum - the problem is, those beverages don't taste great right now. Bartenders don't want to use it in cocktails, and western palates are prone to hedonic rejection (to use the technical term)[1].

What will happen next?

We've already seen the intrusion of traditional European and American Alcohols throughout Asia, and Baijiu will need to evolve to compete to be something that people want to drink. If there are any distillers reading this, reach out to me[2] - we have lots of experience optimizing flavor profile.

[0] www.Gastrograph.com [1] read: "disgust" [2] JasonCEO [at] [the above URL]

This was actually the topic of the closing address at the Master Brewers Association of America (basically a modern guild meeting + trade show).

The main thrust of the talk was that nutritionally, beer and wine should be about equal, but everyone believes that wine is healthy and beer isn't - which is not scientifically accurate.

A more nuanced view: ethanol (alcohol) accounts for nearly all of the cardio-vascular benefits, and thus, beer and wine are equal. Wine contains more anti-oxidants and anti-aging properties, but you would need to get very (very) drunk to get those benefits (well past the Ballmer Peak). Beer, specifically craft beers, have great phytonutrients and macronutrients - and you can get those benefits from a single bottle.

In my opinion: advantage beer.

I have the most mixed feeling about that XKCD.... it's funny, but really misses the mark for people who _enjoy_ craft beer.

I think that the explosion in choices, availability, and quality have been great for consumers, American manufacturing, and the community that these breweries build around themselves.

[full disclosure: I work in the beer industry)

We use our own staff for R&D (training the models), but semi-trained tasters are reviewing the product at the brewery;

we weight individuals on a verity of factors and have a pre-processing method that controls for experience level in expected flavor perception per flaw.

So experience level is important, but we often have to work around it. Most of the employees of a brewery or other beverage manufacturing are relatively experienced anyway!

Hey Sprocket!

We can do any homogeneous product made in batches - so cheese may be a good fit!

We've done some yogurt and cheese reviews in the past to test the system but nothing extensive.

Send me an email at JasonCEO [at] Gastrograph [dot] com and we'll see what we can do?

Cheers! - Jason

The models are written in R and trained on spiked samples of beer - we add a known chemical contaminant to the beer, review the product, and train the models.

Depending on the style of beer, we sometimes use our clients products, sometimes representative a representative beer from the style of interest. Depending on the flaw, sometimes the models are global (style independent) and sometimes there's a model per style.

We do some marketing in trade publications such as New Brewer and Roast magazine (for coffee). We also attend trade shows.

Any marketing ideas?

Hi gballard!

The tastings are done in-house by the companies production team. There are no samples to ship, and the sensory system is easy enough that just about anyone in the industry picks it up there first 4 - 8 times (its often more in-depth then they're used to!).

Thank you for the kind words!

We have a couple thousand reviews per style, so its really fat-data not big data. (in the tens of thousands for some styles)

The main difference is that we take a complete sensory profile from every review, and join that with environmental and personal effects, such as time of day, day of week, temperature, altitude, etc and learn the reviewers preferences and sensitivities over time. We take about 600 variables per review in total.

From that data, we're able to generate a much more actuate and actionable analysis (at a cheaper cost) than a GC/MS or HPLC could - just consider, your mouth is a perfect tool for tasting the compounds that matter to you when deciding what you like and dislike. This is modified by your environment, and we need to control for how your perception of flavor will change over time with age, exposure, and experience.

I believe that a GC is a waste of money for 90% of breweries.

Wow, 2 articles published about us in a day!

also being discussed here: https://news.ycombinator.com/item?id=10159852

Hello fellow HNers,

I'm the CEO of Analytical Flavor Systems!

Ask me anything!

- Jason

For anyone interested in beer and data science, my startup[1] uses machine learning and artificial intelligence to build flavor profiling and quality control tools for craft beverage producers.

Our models flag and predict flaws, taints, contaminations, and batch-to-batch deviations in real time from human sensory data. We then leverage our clients quality control data for flavor profile optimization, demographic targeting, and cognitive marketing - helping them sell consistently better products to their most valuable consumers.

[1] www.Gastrograph.com

Interesting take. I find the categorization around 'other foods' interesting, if not particularly novel. If there's anyone hanging around this board interested in applying machine learning and data science techniques to human sensory data... my company[1] builds flavor profiling and quality control tools for the craft beverage industry. Hit me up[2]!

[1] www.Gastrograph.com [2] JasonCEC [at] ^above url

Taste is an incredibly difficult thing to measure and understand; my startup[1] builds machine learning based tools for flavor profiling and quality control in the craft beverage industry - I'm speaking from experience!

Before you can even approach understanding what a specific product tastes like, and what individuals and populations like or dislike about the product, you need to:

1. Develop a consistent system for collecting flavor profiles - while eliminating exogenous factors such as unattributed variables, halo and masking effect, environmental effects, and cognitive effects.

2. Collect thousands of reviews. Either from an IID sample of the population (sure...), or include the collection of demographic and personal identifiers within the data.

3. Understand the effects of age, race, sex, socio-economic status, and past tasting experiences.

4. Build basic flavor models, test that your system isn't leaving out important information that causes unseen variance within the product reviews, and continue collecting data.

After all of that, the real fun begins! At this point, we can determine most of the chemistry of a product, its risk of flaws taints and contaminations, its production process, and its optimal target demographic just from a few full-sensory reviews.

But we don't rely on standard sensory science to do it - that's stuck trying to figure out how Craft Foods can add less of a cheaper sugar substitute before people notice.

[1] www.gastrograph.com

If anyone is interested in a start-up[1] that works with beer, shoot me an email[2]! My company builds flavor profiling and quality control tools for craft beverage producers using machine learning, sensory science, and analytical chemistry.

[1] www.gastrograph.com

[2] JasonCEC [at] gastrograph [dot] com

-------- edit / follow-up -------

I believe that the growth in artisan firms is part of the trend towards affordable luxury goods, where individuals who find products that better match their preferences at nearly the same cost (not price^), become brand or product loyalists - they'll continue purchasing and consuming that product until it fails them (quality control) or their preferences change (which happens quite often, usually as a result of experience - we build flavor profiling tools to track and re-target these individuals).

^ example: the average craft beer may cost ~2x as much as Bud or Natural Light, but contains >2x alcohol by volume.

For anyone interested in food and technology, my company[1] builds quality control tools for craft producers. We use machine learning and data science to help producers refine their production process and understand what specific populations and demographics like and dislike about the flavor profile.

[1] www.gastrograph.com

For anyone interested in a startup working in the craft beverage industry, my company builds quality control and flavor profiling tools using machines learning. We focus on the craft beer, artisan coffee, and premium spirit industries.

www.gastrograph.com JasonCEC [at] the above url

No, Netflix uses collaborative filters (not feature learning) for recommending shows, and does not collect feature data on individual "consumer" preference (at least not publicly).

What my company does uses sensory and preference data at the individual level, and projects for demographics and populations. Good food doesn't feel fake. Good drinks don't feel fake. In the long run, few products will be made by averaging tastes for the mass market - that's what coke and pepsi are.

My company[1] is a similar service for the artisan/premium beverage industry (focusing on coffee, beer, and spirits). We use machine learning and data science to understand what individuals and populations taste in a product, what attributes they're selecting for, and what they like and dislike in the product.

We then use that data to monitor their products in real time for quality control problems (flaws, taints, contamination, and batch variations), and create production and flavor profile optimizations.

I think Shazam (and my company) are offering a valuable service to artists and creators, who never had access to these types of optimizing and feedback tools before. At some level, music and cooking is an art - but its also a business, and the art is based on appreciation and consumption; and thus feedback, assessment, and optimization are critical.

I believe in the future, both music and high end drinks will be created on a near individual level - that they will be targeted and modified and optimized for very small groups and demographics. I believe that is a good thing for everyone; better products, higher quality, made for me, and that its tools like [1] and Shazam that are making this possible.

[1] www.Gastrograph.com

Please share your experience with active inside sales (when you have some anecdata) - I know many of us would find it a valuable read.