.png&w=640&q=75)
.png&w=640&q=75)
.png&w=640&q=75)
🤓 Latest Submissions
VLA-Powered LEGO Quality Inspection Robot
Our system combines SmolVLA and ACT (Action Chunking with Transformers) to learn robot movements from a human-operated leader arm. Using Physical AI, we collect real-world demonstration data by guiding the robot through LEGO inspection and sorting tasks. The robot learns to identify LEGO blocks, determine whether each piece is perfect or defective, and autonomously pick and place it onto the corresponding plate. We use an H100 GPU for imitation-learning training, enabling the model to learn manipulation policies from the collected demonstrations. To make the system easier to understand and troubleshoot, we also built a debugging and visualization tool on Rerun.io, allowing us to replay demonstrations, inspect robot actions, and analyze model behavior step by step. Pipeline: Human Leader Arm → Physical AI Data Collection → SmolVLA + ACT → H100 Imitation Training → Robot Pick & Place → Perfect / Defective Plate → Rerun.io Debug & Replay
16 Sep 2026
PrivateDoc
PrivateDoc is an offline, on-device document intelligence assistant designed for sensitive personal paperwork, running on a Samsung S25 Ultra phone powered by Qualcomm SnapDragon. It combines ML Kit OCR, ExecuTorch, MiniLM semantic retrieval, and an on-device Llama 3.2 1B LLM to scan, classify, index, and answer questions about prescriptions, contracts, bills, IDs, and other documents without requiring an Internet connection. A strict retrieval-augmented generation (RAG) pipeline ensures every response is grounded in retrieved OCR evidence, with supporting text highlighted on the original document to improve transparency and reduce hallucinations. By performing OCR, retrieval, and language generation entirely on the device, PrivateDoc preserves user privacy while delivering fast, reliable, and source-backed document understanding.
28 Jun 2026
PlanMyDay
It uses the bright data MCP and scrape/SERP APIs to find activities in the bay area across a few dozen sources. Parents of kids, people looking for fun activities, or even tourists who want to tour like a local can browse through it and find the most interested local activities. The tool also provides AI-generated summary of events as well as the items to bring to the activites. It also features a one-stop solution to let AI find out the activities and lay out the plan for your time period. There system can be automated with frequent scraping and maintenance of the data to keep track of the developments of the events. In the future, I hope to expand it into more localities so more people can use it for activity planning rather than having to browse many different resources. This tool can also attract event organizers to promote its events and for advertisers to make use of it.
31 May 2026
