We present details of the Automated Radio Telescope Imaging Pipeline (ARTIP) and results of a sensitive blind search for HI and OH absorbers at $z<0.4$ and $z<0.7$, respectively. ARTIP is written in Python 3.6, extensively uses the Common Astronomy Software Application (CASA) tools and tasks, and is designed to enable the geographically-distributed MeerKAT Absorption Line Survey (MALS) team to collaboratively process large volumes of radio interferometric data. We apply it to the first MALS dataset obtained using the 64-dish MeerKAT radio telescope and 32K channel mode of the correlator. With merely 40 minutes on target, we present the most sensitive spectrum of PKS1830-211 ever obtained and characterize the known HI ($z=0.19$) and OH ($z=0.89$) absorbers. We further demonstrate ARTIP's capabilities to handle realistic observing scenarios by applying it to a sample of 72 bright radio sources observed with the upgraded Giant Metrewave Radio Telescope (uGMRT) to blindly search for HI and OH absorbers. We estimate the numbers of HI and OH absorbers per unit redshift to be $n_{21}(z\sim0.18)<$0.14 and $n_{\rm OH}(z\sim0.40)<$0.12, respectively, and constrain the cold gas covering factor of galaxies at large impact parameters (50 kpc $<\rho<$ 150 kpc) to be less than 0.022. Due to the small redshift path, $\Delta z\sim$13 for HI with column density$>5.4\times10^{19}$ cm$^{-2}$, the survey has probed only the outskirts of star-forming galaxies at $\rho>30$ kpc. MALS with the expected $\Delta z\sim10^{3-4}$ will overcome this limitation and provide stringent constraints on the cold gas fraction of galaxies in diverse environments over $0<z<1.5$.

We present two algorithms to identify and flag radio frequency interference (RFI) in radio interferometric imaging data. The first algorithm utilizes the redundancy of visibilities inside a UV cell in the visibility plane to identify corrupted data, while varying the detection threshold in accordance with the observed reduction in noise with radial UV distance. In the second algorithm, we propose a scheme to detect faint RFI in the visibility time-channel plane of baselines. The efficacy of identifying RFI in the residual visibilities is reduced by the presence of ripples due to inaccurate subtraction of the strongest sources. This can be due to several reasons including primary beam asymmetries and other direction dependent calibration errors. We eliminated these ripples by clipping the corresponding peaks in the associated Fourier plane. RFI was detected in the ripple-free time-channel plane but was flagged in the original visibilities. Application of these two algorithms to 5 different 150 MHz datasets from the GMRT resulted in a reduction in image noise of 20-50% throughout the field along with a reduction in systematics and a corresponding increase in the number of detected sources. However, on comparing the mean flux densities before and after flagging RFI we find a differential change with the fainter sources ($25\sigma <$ S $< 100$ mJy) showing a change of -6% to +1% relative to the stronger sources (S $>$ 100 mJy). We are unable to explain this effect but it could be related to the CLEAN bias known for interferometers.