> abhay kashyap_

This was built as a proof-of-concept for Voice User Interfaces (VUIs) -- driven by my deep desire to eliminate typing.

It started as a personalized shopping assistant for mobile, with a purely voice-controlled interface. Later, I brought that to my couch as a native TV app, blending remote-as-a-mouse with voice-as-a-keyboard.

I built it in early 2025 with Pipecat, when the tooling landscape was quite primitive. I also demoed this at AI Tinkerers.

This started as a simple Streamlit app to download and transcribe YouTube videos, right after Whisper launched. After that, I extended it into a knowledge layer for RAG, with tooling to embed, index & search across videos.

I gave a short talk about this at AI Tinkerers & TwelveLabs. It picked up 750+ stars on GitHub before I moved on.

Plumage was an early exploration of AI-generated fashion design across apparel, home goods, and tattoos.

It combined object detection with Stable Diffusion's inpainting, and was deployed as a platform to create, share, and rate designs.

Me and my buddies, Xi & Reza, hacked this up in a couple of weekends with OpenAI's CLIP, Pinecone & Django.

We hoped for Lambos to magically show up after launch but they didn't so until then, ape together strong 🦍 🦍 🦍

WildePod is an AI-powered, human-in-the-loop system to manage camera trap images and annotations -- built for Felidae Conservation Fund, a wildlife conservation & research organization.

I built this back in Fall '21 using Django, MegaDetector, Annotorious -- it was also my first major AI-assisted coding project with GitHub Copilot.

Rapalytics let you explore and compare lyrical features across rap artists -- built off 10k+ songs from 150+ artists with D3 & Django.

It's also been cited in a couple of academic publications [1][2].

It accompanied my PhD research at the time & I haven't maintained it since, so it's dated.

Semantic Textual Similarity (STS) measures how close two sentences are in meaning. A system scores each sentence pair, and those scores are graded against human ratings. For much of the 2010s it was a key benchmark for sentence embeddings & large language models. It ran as a SemEval / *SEM shared task, where our lab competed in 2013 and 2014.

Our 2013 system used LSA-based word embeddings fused with WordNet and ranked best among 89 submissions. In 2014, we extended it to handle named entities via knowledge graphs, multiple languages, and text of different lengths, ranking first in several of the subtasks.

This became a 2015 journal paper and a core pillar of my PhD dissertation, where I tackled cold-start & long-tail for music recommendation using deep neural networks.