I agree with those who recommend starting by working on one's derivational morphology.  This page

http://conlang.wikia.com/wiki/List_of_derivation_methods

might help, more in the stuff it links to (also the mailing list links on the Talk: page) to than in its actual content, so far.

I would also second the idea of working on one semantic domain at a time, especially in the early stages.  In particular, this lets you think at length about how you or the hypothetical speakers of your conlang would conceptualize those things, at what perceived fault lines in the continuous mass of reality they would cut their ideas into words.

Someone complained about the vagueness and polysemy of the glosses in the Swadesh list, particularly prepositions like "at".  For the semantic field of adpositions, locative cases, and so forth, you might find this old CONLANG list thread helpful:

http://www.pobox.com/~jimhenry/conlang/prep.txt

The specific analysis methods proposed there shouldn't be implemented directly except in an engelang or auxlang, of course, but they can be fruitful of ideas for naturalistic artlangs.  English has distinct "on" and "onto" and "above" but conflates all those concepts for the downward direction as "under"; maybe your people would do the reverse, or might make similar distinctions for "left" and "right"?

As for continuous development of a long-term conlang project, the main thing is, as an earlier poster (tsiasuk-pron?) 

There have been a couple of recent threads on the CONLANG list that are relevant to this topic: this one, on classification systems, might give you ideas about semantic domains to work on one at a time,

http://archives.conlang.info/qhau/kenwia/tree.html

This one is about ways to avoid coining words too similar to ones you've already got (either absolutely, or within a given domain):

http://archives.conlang.info/fau/weildhon/phuerphuetal.html


or Daniel Feuchard?  The addresses I have for them are bouncing.

--
Jim Henry
http://www.pobox.com/~jimhenry/conlang/fluency-survey.html
Conlang fluency survey -- there's still time to participate before
I analyze the results and write the article