Yatin Kumar

Stock Signal App- LLM-Driven Market Intelligence Pipeline

Cut end-to-end inference latency to ~2–4s by parallelizing ingestion with AsyncIO and batching FinBERT sentiment inference.

Quantixiom Chatbot - Production GenAI Integration

Architected a plug-and-play provider-adapter pattern across OpenAI, NVIDIA NIM, and Ollama, cutting provider-switching integration overhead 70%.

Thera-Mistral - QLoRA Fine-Tuned Conversational LLM

Fine-tuned Mistral-7B via QLoRA on a CBT-informed dataset, training 41.9M adapter parameters with 4-bit NF4 quantization, reducing VRAM requirements 65% (14GB to 4–5GB) on consumer GPU hardware.