Slim hybrid coding-agent harness: mini-swe-agent loop with tau-style tools, multi-provider support, skills, guardrails, and Rich TUI.
Technologies Used:
Self-hosted AI team workspace: named LLM bots join chat channels, take jobs, run tools on your machine, hand off work, and leave an audit trail.
Technologies Used:
Multi-agent AI system for math, physics, and chemistry problems using LangGraph orchestration with parallel specialist agents.
Technologies Used:
Agent-agnostic local LLM evaluation and regression testing gate with DeepEval, golden datasets, SQLite baselines, and CI-ready exit codes.
Technologies Used:
Fine-tune LLMs on your GPU. CLI tool for QLoRA fine-tuning, DPO alignment, and eval — concurrent jobs, auto-queuing, live metrics.
Technologies Used:
Semantic LLM router that classifies prompts into cheap/mid/hard and routes to Groq, OpenRouter, or your upstream. OpenAI/Anthropic-compatible API, MCP server, and chat demo.
Technologies Used:
RAG system that remembers. Answers from episodic memory in ~260ms instead of ~49s.
Technologies Used:
SIFT is an autonomous AI research agent using LLM tool calling, web search, and a dynamic reasoning loop. Built with FastAPI, Streamlit, and OpenAI-compatible APIs.
Technologies Used:
An end-to-end ML pipeline for classifying anime characters. Features Optuna hyperparameter optimization and a two-stage training strategy (Linear Probing + Fine-Tuning) on a ResNet34 backbone.
Technologies Used:
Predict if a Spotify track will chart-top or stay underground using audio features like energy, danceability, and valence.
Technologies Used: