The article was a bit outdated since a couple of the used semiconductors went out of production already so I ran into this site here  by DK1RM, Roland, who created an Eagle layout and updated the circuit design with more current semiconductors.

His site publishes the schematics too. 

What the generator does is to use a standard TTL oscillator, attenuate its output signal with a FET, pass it through a filter to achieve a sinus signal. The sinus signal is now fed into a peak detector and the peak voltage is compared to a DC voltage out of a voltage regulator. The output of the comparing opamp is then used to control the FET attenuation, creating a control loop, delivering a sine signal with fairly constant and defined amplitude. This is then passed through fixed attenuators which in turn generate output levels at 0 dbm and -60 dbm.

I have had the layout that Roland publishes on the website manufactured and built the resulting device into a RF tight enclosure. To my surprise the resulting 0 dbm output was as precise as my instruments could capture but not entirely unexpected (Roland writes about this on his site), getting the -60dbm output to deliver a precise level is a serious pain in the neck. For that reason I have decided to follow his advice to change the attenuator to -50dbm output which works more or less fine for me.

Unfortunately his web page is in German so you will want to use a translation service. Here is a link that does that. 

Let me know if you require help with understanding the content.

Of course it might be interesting to use a precision DC reference for comparison with the peak detector signal but the results with this circuit have so far shown to be accurate enough for amateur purposes. The 0dbm output in combination with a commercial (used) variable precision attenuator (look on eBay) give an awesome instrument to calibrate against.