0:01

Hello, and welcome, everybody. Guys, it is getting close to

0:04

Christmas. This is Joe from Sudapri. Io, and therefore, I'm bringing

0:08

you a very special bonus episode this week

0:12

to you very shortly before Christmas. But nonetheless, I would

0:15

like to welcome Daniel here. Hey. How are you doing?

0:19

Hey, Jan. I'm doing well. Thanks. Thanks for having me.

0:23

AI pleasure. We may tell our audience that this, recording

0:27

is sponsored by Frankfurt, meaning the business development

0:30

agency who is also supporting Frankfurt Forward. And

0:34

the reason you are here, you guys won start up of the

0:38

year 2024. Congratulations

0:42

to that. Did

0:45

this recognition first, can

0:49

you tell us a little bit about you and your company

0:53

before we get into the specific questions? Yeah.

0:56

Sure. Yeah. So

1:00

Daniel Iglesias is my name. I am from the AI Main

1:04

region. I AI married to teacher. My

1:09

professional background is in banking technology consulting.

1:13

That's what I did before I founded,

1:17

Digisapiens in, 2020.

1:22

And, what we do at Digi DigiZapiens is

1:26

we produce speech recognition systems

1:30

that not just recognize what is being

1:33

said, rather how it's being said. So

1:37

it's a speech recognition system that is geared towards

1:41

special use cases, that

1:45

are relevant to measure how well somebody speaks or

1:48

reads. And with the help of this speech

1:52

recognition technology that, is AI,

1:56

so that we've built ourselves, you can build

2:00

use cases in especially in the area of

2:04

education. So that's where we started. And the first use

2:07

case that was built with our speech recognition technology

2:11

was the LALI 2. LALI 2

2:15

stands for loud laser tutor in German, which is,

2:19

reading aloud tutors. So it's it's a tool,

2:22

that listens to students in schools

2:26

from 2nd to 7th grade in Germany.

2:30

While they read aloud, it analyzes how

2:34

well they do that, so it carries out a diagnosis.

2:39

And after that diagnosis, they are,

2:43

being trained to become better readers

2:47

based on the, diagnosis that was carried out before

2:50

that. So basically, AI looking for a question here. Sure.

2:54

When you said they are your tool

2:58

helps, how well they are speaking

3:01

their AI the tone of voice recognition. Is it only

3:05

how well they vocalize the tones? Or is

3:09

it also that you can deduce some some

3:13

level of their understanding of what they're reading?

3:18

The speech recognition technology itself is,

3:23

audio technology, so it listens. So we it detects everything

3:26

that can be caught by a microphone.

3:30

And, in contrast to, I

3:34

call it, regular speech recognition systems where the goal is to

3:37

detect the probable intent. So what would

3:41

be this what are you probably meaning? So,

3:45

the goal is to carry out a a

3:48

task or, AI I'm

3:52

looking, a command. You have to carry out a command. Mhmm. Yeah.

3:56

Play this and that song or whatever. In contrast to

4:00

that, we really transcribe and list listen to what has

4:04

actually been said. So this includes

4:07

arrows AI my that I already said now a few

4:11

times. And repetitions, or

4:15

text so regarding text repetitions, things that have

4:18

been left out or added, the tonality

4:22

of things, and also whether you pronounce words

4:26

correctly. Yeah. So that's what the, speech

4:29

recognition technology does. But we also develop,

4:33

systems that help understand help

4:37

students understand the text better by

4:41

generating, differentiated quizzes. So, you cannot ask

4:48

every reader the same question. It must be adapted to

4:52

his reading level and also understanding

4:56

capabilities. So the complexity of the questions

5:00

and the possible answers also must be adapted. So,

5:04

all in all, we're in the business of providing

5:10

education, specialists, with,

5:14

with the necessary tools to build very innovative

5:17

adaptive tools, for learning

5:22

reading skills or language skills? That is exactly what I had in mind.

5:29

Vividly remember when, for the first time

5:33

in my life, I understood a Chinese joke

5:36

about foreigners, instead of how

5:40

how, she said, how how. So the the 1 means

5:44

good good, the other means mouse mouse, different tones. So the the

5:47

the the question is here, how

5:51

many languages can you do? And isn't something like

5:55

Chinese where there's a different if I ask or

6:00

the 1 we means please ask. So if you ask for direction,

6:04

the other one's the other 1 means please kiss. Well, I made

6:07

an older Chinese lady on the streets of Beijing really blush.

6:12

How many of those differences could you actually do?

6:15

Because on the top of my mind, yes, of course, English is a little bit

6:19

difficult to pronounce Spanish as well. But if the tone really makes a

6:22

difference, like in languages, like Chinese, Cantonese, and so on and so forth,

6:27

that would weigh your AI way 2 would really come

6:30

in handy. So my question would be, how

6:34

much can you do there? Could you, give us an

6:37

idea of the granularity and languages you cover? Yes.

6:41

We so as of now, we're covering German and English.

6:46

We did not experiment with Chinese since the

6:49

Chinese education technology market is highly regulated

6:53

and basically close towards foreigners.

6:58

But, we my team, I have

7:02

dedicated experts in,

7:05

working with non Roman

7:09

languages, so especially also Indian languages.

7:12

There's 40 something. Sorry that I cannot recall

7:16

the exact number, but there's, more than 40

7:19

languages, official languages spoken in India. We can work with

7:23

those. We can also work with, Arabic

7:28

character sets. But as of now, in terms of

7:33

solutions that are at hand, we can work with

7:37

Germany, German language, and, we

7:41

or next year, we will also launch, the English version

7:45

AI API, and other languages.

7:49

We have the skills and capabilities to train and

7:53

fine tune models, with a short pilot

7:56

project that we need to carry out with potential customers.

8:01

So that means you already, by the way, I linked it in the show

8:05

notes. La Lalu. Who is it?

8:08

La la la la la la la la la la la la la la la

8:12

la la la la la la la la la la la la la la la la la. It's There's a there's a, a

8:18

song that you sing to children before they go to bed,

8:22

AI. It's called. Yeah?

8:26

And, we have some similarities there. So it's called

8:30

too. It's about 3. Mhmm.

8:35

So, sorry. Just

8:39

typing here, that we also linked the

8:43

song here, in the show notes. My my quest

8:47

so this already establishes something we

8:51

could deduce from what you're saying. So, basically, you are

8:55

not a customer facing product. You're 1 of the tools, the APIs,

9:00

others could include, could work with in

9:03

developing their own client facing b to b, b to c, b to

9:07

g tools. Right? Yes.

9:10

With the asterisk, we we have

9:14

developed the LALI 2, for our partner at hence,

9:18

So we do develop platforms and applications,

9:22

but we rather license our the speech technology

9:27

to partners, b to b, b to g, b to c, whatever.

9:31

We are already discussing, the tool. But

9:34

before that, I I actually wanted to be because I have so

9:38

many questions. Before that, I would actually, wanted

9:42

to ask you

9:46

where this idea is coming from. I do believe I have an idea since you're

9:49

married to your teacher, but you have been working in banking,

9:53

finance, technology, triangle consulting. So

9:57

so, where did the idea come from? How did you get

10:01

that? And especially the question, when did you decide

10:04

jump ship to really do this full time?

10:08

Okay. So I was, like you said, I

10:12

was about 17 or 18 years into banking technology,

10:16

and, I always had the goal

10:20

to promote young people

10:23

in achieving, how do you call

10:27

it, higher education. Let's put it this way, to

10:31

get the most out of their potential. So I did,

10:34

trainings in schools for how to apply to a job,

10:38

how do I choose the right job for me, etcetera etcetera. And

10:42

I always had this passion for helping young people.

10:45

So, that's that that passion was always there. But in

10:49

2019, that was the time where I really,

10:55

thought about what can I do with the skills that I have and the

10:58

knowledge that I have? And, to to really have an

11:02

impact on our youth in a bigger scale. Because

11:06

this is the AI, shortly after AI became a father of

11:10

a of a daughter, and I have observed

11:13

certain trends in our society in Germany. 40 to

11:17

50% of the children have

11:21

background, with non German parents or migration

11:25

background, as you would call it in Germany. And this

11:28

leads to some hurdles and

11:32

some difficulties, in,

11:35

in the school system, and and and since

11:39

we also, at the same time, have a shortage of

11:43

shortage of teachers. And I have also

11:46

observed what happens in the market regarding

11:50

the upcoming AI of AI, robotics, automation.

11:54

I was part of it in banking. So I added

11:59

the deterioration of reading skills

12:03

towards higher requirements

12:07

regarding job skills and came up,

12:11

with, with a

12:14

perspective that I didn't like for the future of my children. So that's

12:18

why I decided to take my skills,

12:22

which is general management, business

12:25

development, technology understanding skills, and

12:30

work together with the best experts I can find,

12:33

in terms of reading capabilities

12:37

and, reading training and,

12:41

excellent techno technology experts bring those things together and

12:45

build what you find today. Mhmm. The tone of voice recognition

12:52

is fascinating. You already told

12:55

us the, loud reading tutor is

12:59

something you develop for a customer. Could you also

13:03

share another example already where an external,

13:08

client is using your tool? The

13:12

second example is in the making. It's not ready to be

13:15

shared publicly, but, I must openly say the

13:19

past was, highly,

13:23

we were highly invested into building

13:26

that tool, which is 1 of a kind. It took a lot of

13:30

attention and all of our resources to get it running in

13:34

time and, make it

13:38

scalable and stable and user friendly. So

13:41

now that the product is fully marketable and

13:45

most of the almost all bugs are fixed, yeah,

13:50

now we are able and ready to focus in on

13:53

new, projects and partners.

14:00

AI see. Talked about

14:03

partners here and winning new clients. Winning the Frankfurt Forward award

14:07

is a huge accomplishment. What do you think made DigiSapiens

14:11

stand out among the competition this year?

14:16

AI think it's the social impact,

14:20

dimension of what we're doing. We are for profit social enterprise,

14:23

so we are here to do good

14:29

and, earn some money at the same time. And I think the audience

14:32

AI that idea, and I think there's a lot of ad

14:36

techs out there, but pulling it off in the way

14:40

that we did, by partnering with such a,

14:45

renowned brand AI Ernst Kedfalak,

14:49

as a first initial project. And at the same time,

14:53

building such a unique technology like we do.

14:57

I think that's what impressed the jury and,

15:02

caused a lot of, support in the audience.

15:09

I see. I see. Your

15:12

technology has potential across industries.

15:17

We already know you're working and focused

15:21

on the education, but could you also see some other

15:25

industries where you could, like, in the future, a few years down the road,

15:29

apply it? Yes. Outside of, education,

15:37

there is also the entertainment and gaming industry

15:41

that could work with our technology,

15:46

where you could you could use it to build games that,

15:51

based on reading skills, which would

15:55

be some sort of, yeah, educational games at at the same

15:59

time, but you can also use it to,

16:04

to build, how do you call it?

16:08

A a presentation trainer or speech trainer that,

16:12

helps you become a better speaker for public speeches.

16:18

The creativity and opportunities are unlimited,

16:22

but everything that involves the ability

16:25

to carefully listen to what and how things are

16:29

being said. You know? Mhmm.

16:34

I AI was wondering, AI sure you

16:38

thought a lot about potential use cases. Could you share, like,

16:41

the the the the the the most interesting, the most,

16:46

quirky 1 you already came up with?

16:50

No no need that it actually AI, but you thought,

16:54

theoretically, our idea could also be applied too.

16:58

Yes. So we have applied it to reading

17:02

learning or reading promotion, but, very,

17:06

very relevant use cases also in the area

17:10

of language learning. You have seen

17:14

big companies like Bubble, etcetera use it in

17:17

some way. Yeah. And we envision

17:22

other ways that are much more focused on

17:25

dialogues, that our technology could be

17:29

used for to promote language skills or learn a new

17:33

language. So we are able to evaluate how

17:36

things are being pronounced. That's 1 major skill. We can

17:40

analyze, literally what has been said.

17:44

We can analyze 4 ds, all areas

17:47

of, language that we can analyze.

17:51

And I think this is relevant if you want to learn a language

17:55

properly. You have, you have given some example from

17:58

Chinese. If you listen to people

18:02

talking German with all their accents and

18:05

dialects, we also have ways to

18:10

tackle, dialect, dialects because the way you use your mouth, your

18:17

tongue, your teeth. So your

18:21

complete speech apparatus is also something that we can

18:25

derive from the audio signal and combining all

18:28

those, all those measurements

18:33

into a cohesive didactical

18:36

concept is something really unique,

18:40

that we haven't seen so far. Mhmm.

18:45

Going into a little bit different topic because every everybody talks

18:49

about accuracy, like fantasizing AI and

18:52

ethical use of AI. With analyzing of tones, you're

18:56

you're, collecting potentially

19:00

sensitive information or your clients do and you process

19:03

this. How do you, ensure

19:07

the ethical use and the accuracy when handling the sensitive

19:11

data? Yeah. So

19:15

the in Germany, it's always a relevant question whether it's a

19:21

personal personality AI identity related data.

19:25

You know? That's 1 major question. And,

19:29

the the thing is, if voice

19:33

really is such data, you would

19:36

need to have some registry

19:41

of confirmed identities

19:45

that are linked to a voice profile or voice biometrics profile

19:49

to to pose some danger

19:54

to a data leak or whatever. You know? Mhmm.

19:58

And this is not the case and will never be the case. We will

20:02

I well, let's let's not say it will never be the case. I don't know

20:05

what happens in, the year 21 100. But as of now, we don't

20:08

have a voice register, a public 1.

20:12

And, the question is also, even if this existed

20:16

on a government level, the question is also, do

20:20

companies, do other individuals,

20:24

criminals have access to this registry, and can they use it to harm

20:27

you? And the I don't I don't see

20:31

that. When you ask me about ethics

20:35

in my context, we regard the topic of

20:39

ethics in terms of accessibility to

20:43

our solutions. So can somebody from Bavaria

20:46

use it, as the same way as somebody from

20:50

Saxony can use it? And can somebody

20:54

with Turkish or Arab

20:57

accent use it AI somebody from, Hanover

21:01

without any accent? And the answer is yes.

21:05

So we AI to and we put a lot of

21:09

effort into avoiding any biases

21:13

in our speech recognition system by

21:17

training it very profoundly

21:20

with different accents and dialects

21:25

to make sure that it works with every

21:29

user. Yeah. So that's how we look at that.

21:33

So you you you put a lot of effort into that,

21:37

making it possible for everybody to understand. We may tell the

21:41

audience that there are some people who speak very, hefty,

21:45

local accents, not only from Bavaria, but AI,

21:49

Thales, Saxony, and so on and so forth, but also Platych

21:53

in the very north. It's really hard for you to understand when you're from a

21:56

different area. Many Americans will understand,

22:01

will, have an idea when I talk about somebody with a very

22:05

heavy southern draw or something that's also hard to understand. So

22:08

you took care to cover all those

22:12

peoples and not disqualify somebody there. So I do believe there

22:16

was a lot of development work going into.

22:20

What challenges did you face, and how did you overcome this in

22:23

developing digital DigiSapiens, not digital.

22:27

DigiSapiens. Sorry. Yes. So

22:32

yeah. There there were a lot. So which 1 can I

22:37

yeah? So we started as a company that wanted

22:40

to provide speech recognition systems only,

22:44

And then we were suddenly in the position to develop a whole

22:47

platform. So in a short time, we had to

22:51

set up a team that was able to do that, build a

22:55

product team or, UX and

22:59

front end development team around our core technology

23:03

team in a very short time and,

23:08

build a product that fulfills high

23:11

expectations. And this, was a

23:15

real challenge. To be honest, we,

23:19

like a lot of other startups, were

23:23

in the forced to publish a product a

23:27

year ago that was not perfect, far from perfect.

23:31

So we got that feedback in the beginning, but we worked

23:35

really, really hard, with our team, which also includes the first level

23:42

support who's directly in contact with us, our schools,

23:47

that use it and the partners to really get the first

23:51

hand impression of what is

23:55

working well and whatnot, and they are

23:59

very much integrated into our development process.

24:03

And we take we took every feedback very

24:06

seriously. And, yeah, now a

24:10

year after we've launched, I am confident to say that

24:14

we have a very unique, innovative, and highly

24:17

effective, reading promotion solution

24:21

that is, yeah, that we can

24:25

be really proud of. I see.

24:29

The AI world is developing pretty rapidly. You

24:33

are right now, I would say, on the cutting edge of development.

24:38

How do you make sure that you remain

24:42

there as 1 of the top solutions keeping up with the with

24:46

the AI developments? We

24:49

thanks, for that feedback. Yeah. And we and

24:53

I, we really work hard to be seen this

24:56

way. So what we do is we invest a lot a

25:00

lot into r and d. Most of our money goes into r and

25:03

d. We publish papers. We participate

25:07

in international conferences, where we

25:12

also do take over tasks and compete against other teams

25:19

in optimizing models, quantifying models, and

25:23

raising accuracy. And we always, come

25:26

up on top, also, leaving

25:30

huge names behind us. So we,

25:36

we regard this as a sport to develop

25:40

new methods, overcome,

25:44

overcome hurdles

25:47

and, yeah, really try to be on the cutting or beating edge

25:51

when it comes to sophisticated speech recognition

25:55

and NLP solutions. Yes.

25:59

Daniel, I'm sure there will be questions on

26:02

where are the papers. I do have a few suspects where you mentioned something like

26:06

this. You will give me after the official end of the

26:10

interview, you will give me, the link, and I'll post it in the show

26:14

notes. I do have a few certain suspects that always request something

26:17

like this. Yes, Claude from Paris. I'm looking at you. Exactly.

26:22

And then they they can dig through it. So,

26:26

let us go into the very last part of the

26:30

interview because I'm now already bothering you for, like,

26:33

almost 40 minutes in this online meeting and,

26:37

more than 25 minutes in actual interview. So,

26:41

don't worry. There there are only a few more questions left.

26:45

I was wondering winning an award like Frankfurt Forward often

26:49

reflects strong local support. How has the

26:52

Frankfurt ecosystem contributed to your journey so far?

26:56

There are a lot.

27:00

So the our ecosystem is

27:03

very regional. Our investor is regional. Our

27:07

network helped us find our first customer,

27:11

is from the region. I we get a lot of

27:15

recognition and inquiries due to that, price.

27:19

So, it's mostly visibility

27:23

and also recognition. So it's when you approach somebody

27:27

and, he asks you who you are, what you do, and,

27:32

you mentioned the start up of the year world from Frankfurt forward,

27:35

especially in the region. Everybody stops questioning

27:39

whether you are or whether what you are doing is

27:42

sound and makes sense. So,

27:46

the intro and entrees into conversation building partnerships

27:50

is much easier. Mhmm. And

27:54

now only 3 more questions left.

27:57

You are AI now a leader in a very specialized niche in

28:01

AI. But what advice would you offer to other start ups

28:05

looking to innovate in the AI and tech space? Some kind

28:08

of skills, like processes you have learned so far,

28:13

not only taking the, your business ideas from influencers

28:17

on Instagram. Yeah.

28:21

I don't know if I'm the right 1 to give advice. Yeah. I'm not,

28:25

the tech guy in my company. But what I would generally suggest, when

28:29

looking at tech is not looking at a hype or the technology,

28:33

but rather trying to solve real world problems.

28:37

So AI in Germany, we have 25% of the children,

28:42

that at the age of 15 do not understand what they

28:45

read, and they don't understand what they read because they are not fluent

28:49

in reading. So that's a huge societal

28:53

problem, and that's, that was the initial thing

28:57

that led me to found it found DigiZapiens.

29:02

And I would really not focus on the

29:05

technology or the AI for the purpose of the technology in

29:09

AI, but rather using tech as a

29:13

means to solve a real problem. Yeah. And if it takes

29:16

AI, it's fine. If not, not.

29:20

But 1 also has to say in German, there can

29:24

be very difficult,

29:28

sentences and structures. III

29:32

everybody who's who tried either in German or in a

29:35

translated version to read Kafka should know what we're

29:39

talking about here. So, therefore, it

29:43

can be difficult. There's a saying in German,

29:47

German language, difficult language. So, actually,

29:51

it it it says good things about you guys that you started

29:54

with the German language and mastered it for your, for

29:58

your, tool, DigiSapiens.

30:03

We usually close out with 2 more questions, and they're usually

30:06

pretty simple and usually end with a yes. But I'll ask them

30:10

anyway. Are you open to talk to new investors?

30:16

And as always, I'll link your LinkedIn profile down here in the show

30:20

notes wherever you're looking this wherever you're watching this. No.

30:24

Sorry. This AI, no watching. But,

30:28

anyways, you you either directly in your tool. I'm

30:31

sorry. Not every tool allows links you can then click.

30:35

So, basically, you could go to our blog, standard break.

30:39

Ioforward/block, and there we link Daniel's

30:43

profile. Plus, when you are

30:47

expanding, when you're growing as a young company, I am sure you're

30:50

all so open to have applications from potential new

30:54

employees. Right? Yes. Yes. Yes. Yes. Yes. We're

30:57

looking for we're looking for good people always.

31:01

AI, is there a career website that I could launch,

31:05

or should the people simply, that I could link, or should the

31:09

people, simply reach out to you via email?

31:13

Yeah. The latter. Directly reach out to us. We,

31:17

we could, we we have a way to go regarding an

31:21

HR department, so everything's handled by the team, depending

31:25

on the competency somebody's applying for. So

31:28

can't use the public channels or directly contact me, and I will forward

31:32

it to the colleagues. Again, go to the blog to the LinkedIn

31:36

profile. Guys, it was a

31:39

pleasure talking to you, Daniel. Thank you very much. Thank you very much for

31:43

answering more than 30 minutes difficult questions here and,

31:48

keeping up my stupid interjection interruptions. Thank you very

31:51

much. Thank you. Nice being here.

32:00

Yeah. So and yeah. Thanks for having me here, and, yeah,

32:04

Merry Christmas and a happy New Year to everyone.

32:09

Thank you. Yeah. Merry Christmas. Happy New Year from you as well.