Timely detection of events in financial news can help in the refined prediction of downstream stock prices beyond what is depicted by volume and transactions. In this work, we have introduced NEIC, a lightweight, parameter-efficient transformer architecture tailored for Indian financial news classification into events of interests. The proposed pipeline integrates TextRank-based extractive summarization and selective text augmentation (including synonym replacement, back-translation, and paraphrasing using a LoRA-tuned T5 model) applied only to minority events that happen rarely but change the course of normal stock price prediction. A domain-adapted MiniLM encoder, pre-trained via masked language modeling on financial news, generates contextual embeddings and is fused with TF-IDF features and cosine similarities to anchors that are class label prototypes, forming a tri-branch feature vector processed by a shallow MLP classifier. This semantic anchor fusion strategy enhances label awareness while maintaining low computational overhead. Our model, with only 24.12M parameters, achieves 94% accuracy, 0.78 macro-F1, 94% weighted-F1, and processes 30 articles in under 15 ms, outperforming several larger baselines. The approach demonstrates that tiny LLMs, when enriched with domain adaptation and anchor-driven semantics, can offer both accuracy and deployment efficiency for real-time financial event classification. Apart from this, this work created an Indian Financial News Dataset that affects the stock price the most and is annotated with careful consideration of the market volatility.