Collecting patient history before a medical consultation is repetitive, time-consuming, and leaves clinicians starting from scratch with every patient. This paper presents a clinical pre-consultation chatbot for telehealth platforms that automates the intake step using Qwen2.5-7B-Instruct, a 7.6-billion-parameter opensource language model fine-tuned on the MedDialog dataset with QLoRA. During a session the chatbot collects symptoms, duration, medications, allergies, and relevant history, then produces a structured clinician note. A FAISS-based retrieval layer grounds responses in curated clinical knowledge, and keyword-based safety guardrails block diagnosis requests and escalate emergencies. Evaluation on 36 held-out MedDialog dialogues yields ROUGE-1 = 0.1842 and BLEU = 0.0243; perplexity falls 92.5% from 44.81 to 3.37 after fine-tuning, confirming successful domain adaptation. The system is a functional prototype demonstrating that parameter-efficient fine-tuning on a single consumer GPU can produce a safe, deployable clinical intake assistant.