video1.0<iframe src="https://www.loom.com/embed/15b9105781024c7fbe0f18eea52c9858" frameborder="0" width="1280" height="960" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>9601280Loomhttps://www.loom.com9601280https://cdn.loom.com/sessions/thumbnails/15b9105781024c7fbe0f18eea52c9858-c47c11a47a7992a3.gif291.792This Loom describes completing an AAML intern task at KASTEC across four components. The author built a seven day Drift persona that tracks mood, tone, and average sentiment, using row binning, and noted that triggers were empty due to noise from repeated entities and a persona without triggers. They implemented an intent classifier with five classes using a local TF IDF plus linear regression approach and a MiniLM model quantized to meet a 50 MB threshold, reporting 1.8 ms latency and weak labeling for all classes. For VAL, they weighted 50 percent cuisine cosine similarity, 20 percent emotional weight, and 30 percent recency, and verified natural language inference with contradiction outputs, completing everything within 24 hours.