Many resource-poor and morphologically rich Indian languages struggle to benefit from recent advancements in feature representations in Natural Language Processing (NLP) because of lack of massive, annotated corpus and sturdy benchmarks. Natural Language Generation (NLG) or Textual content Generation is a subfield of NLP. A Markov chain is a stochastic model that forecasts future states based exclusively on the current state, utilizing a random probability distribution. It is computationally efficient, quick to execute, and requires minimal memory because its next state depends exclusively on the present state, with no influence from preceding states. In this paper, we have implemented 1) A word level Text Generation model with a Markov chain and 2) A character level Text Generation model using a Markov chain with the help of reinforcement-learning. We have used the Telugu dataset used in [1] for our experimentations. We generated only 20, 30 and 50 percent of the input text as summary with both our methods. These generated summaries used for text classification and evaluated the results. Our methods proved as computationally efficient.