Accurate indoor positioning is crucial for autonomous agents, particularly in environments where obstacles obstruct direct signal paths, causing reflections and ranging errors. Ultra-wideband (UWB) technology offers high temporal resolution and robustness to multipath interference, making it a promising choice for indoor localization. However, in non-line-of-sight (NLoS) conditions, signal reflections distort distance estimates, degrading positioning accuracy. Channel Impulse Response (CIR) is a key feature of UWB signals that captures the arrival time and strength of multiple signal paths, including direct and reflected components. By analyzing CIR, it is possible to identify and correct ranging errors caused by multipath propagation, enhancing localization accuracy. This paper presents a novel Transformer-based framework that processes CIR data to predict UWB ranging errors and dynamically adjusts a Weighted Least Squares (WLS) positioning algorithm. Unlike conventional WLS, which assumes fixed measurement weights, our approach learns adaptive weights based on predicted errors, significantly improving accuracy. To our knowledge, this is the first work leveraging Transformers for UWB CIR-based error estimation in WLS. We validate our method on a public dataset covering four indoor environments—residential, confined residential, industrial, and office—each with both clear and obstructed line-of-sight conditions. Results show that our Transformer-WLS approach outperforms traditional WLS and is competitive with CNN-based models. In a challenging industrial setting with severe NLoS, our method reduces mean positioning error from 4.00 m (CNN-WLS) and 1.33 m (WLS) to 1.08 m. Statistical analysis confirms the significance of these improvements ( \(p < 0.01\) ). These findings highlight the advantages of Transformers in mitigating multipath effects, enhancing UWB-based indoor positioning, and providing a scalable solution for complex environments.