This is a Phi Family of SLMs book for getting started with Phi Models. Phi a family of open sourced AI models developed by Microsoft. Phi models are the most capable and cost-effective small language models (SLMs) available, outperforming models of the same size and next size up across a variety of language, reasoning, coding, and math benchmarks

Pivotal Token Search

🔧Fine-tune Phi-4-mini-instruct model with LoRA on Azure Machine Learning Studio

A Telephonic Ai Agent,Listen-Think-Speak in Malayalam,Using Local LLM Stack

A local, privacy-preserving auto-grader built with LangGraph and Ollama. It evaluates student submissions against a JSON rubric using a pipeline of specialised agents, producing a structured Markdown feedback report — entirely on your own hardware with no external API calls.

How to Build a Local Agent

An ablation study adapting 4B-parameter LLMs (Qwen-2.5, Gemma-3, Phi-4) to the Indian Legal Domain. Features LoRA/QLoRA optimization, custom synthetic data generation, and an automated LLM-as-a-Judge evaluation pipeline.

Offline multimodal AI system with speech, vision, and language processing. LangGraph orchestration, Phi-4-mini LLM, Vosk STT, BLIP vision. Complete privacy, zero data transmission. Inspired by Interstellar's TARS.

RAG AI that uses Phi-4:3.8b Mini 4K and All-MiniLM locally

An AI Agent to enable a "chat" with your QIF files.

This is a Phi Family of SLMs book for getting started with Phi Models. Phi a family of open sourced AI models developed by Microsoft. Phi models are the most capable and cost-effective small language models (SLMs) available, outperforming models of the same size and next size up across a variety of language, reasoning, coding, and math benchmarks

A multitask framework for joint ASR and perceptual attribute analysis of dysarthric speech (ASRU 2025).

DejaFood é um projeto que envolve várias camadas de AI, desde a detecção e reconhecimento de alimentos por imagem, geração de receitas e busca na internet de receitas.

Self-hosted LLM security engine for SOC teams. Use any Ollama model to analyze threats, classify attacks, and score risk - fully offline, no cloud APIs, your data never leaves your network.