Peter Preketes Claim this profile

Python 3 and PyTorch 2 revival of DeepTTE for learning and benchmarking neural travel-time estimation

Event-driven CVE remediation that turns security findings into validated pull requests using the Devin API

Vector database built from scratch with five search algorithms, quantization, and RAG integration

AI voice assistant with real-time conversation, document RAG, long-term memory, web search, and interview practice

Real-time Old School RuneScape object detection using YOLOv4, OpenCV, and a custom training dataset

Local book-scale RAG system with hybrid search, Qdrant, Ollama, and source citations

Learned LLM router: predicts per-query whether a cheap model suffices, with honest cost/quality tradeoff curves. WIP.

Curious FDE examples for PII detection, hashing, tokenization, and keeping private business data outside LLM prompts.

CLI for indexing, searching, summarizing, and asking questions about video with TwelveLabs

Fork of Apache Superset, an open-source platform for data exploration and visualization

Working GPT-4 notebooks covering few-shot prompting, self-consistency, ReAct, and iterative refinement

Personal site for notes and the main themes behind what I am building

Fork of Outfit98, a Windows 98-style outfit picker with clothing uploads and AI-generated previews

Working Python examples for OpenAI API patterns including function calling, embeddings, assistants, and structured outputs

Self-hosted n8n workflows for AI image generation, research digests, and local RAG automation

Fork of MicroNxt, an open-source Old School RuneScape client

Multi-agent content generation with Gemini and Griptape, coordinating researcher and writer agents

Automated free-to-play Runecrafting for Microbot with progressive rune unlocking and event-driven state management

Example RuneLite plugin authentication using license keys and hardware-bound identity checks

Tailors resumes to job descriptions with Claude or Gemini and exports ATS-friendly PDFs

Generates consistent image captions with GPT-5 Mini for faster, cleaner LoRA training datasets

Transcribes medical audio with Whisper and uses Claude to extract clinical entities, PHI, and structured summaries