The whitespace after a let-binding is currently optional, this allows expressions like the following:

let x = 3let y = x in y

let x = NaNin x

The reference implementation accepts these expressions while the dhall-haskell rejects them.
Surprisingly

let x = Truein x

is also rejected by the reference implementation.
To avoid such inconsistencies and confusing expressions, I'd suggest changing the dhall grammar to require whitespace after the let-binding.