Portfolio
Full Stack Project
flightviz.net - Video Transcript Indexing
Rather than watch a 2 hour video or have to fast forward through it to get to the part you want, I built an API in Rust to return the timestamp based on the exact search query. The transcripts were parsed and stored in a Postgres Database
Open Source
jobstr.work - NOSTR Job Listings
As a proof of concept I built this in Rust using the NOSTR SDK to fetch messages containing #jobstr. The protocol is based on very simple & flexible event objects (which are passed around as plain JSON) and uses standard elliptic-curve cryptography for keys and signing.
AI & LLM
Semantic Search
Video Transcripts made searchable by semantic meaning rather than keywords. Search with intent! Using an open source Large Language Model and Qdrant Vector Databse I built this with a Python and Rust back end and built the API enpoints using Robyn. I configured Nginx to act as the proxy.
Image Classification
Detecting a human face
In image recognition, embeddings help to distil the essence of an image into a format that computers can analyze and compare - I used Qdrant vector database and FastEmbed to create the embeddings. Qdrant can handle millions of vectors without significant degradation in performance.
Vector Search
Real Estate Semantic Search
Using Python to extract and parse the data. Rust and Qdrant to do embeddings using AllMiniLML6V2 LLM. Full documentation generated by Rust (Cargo). Use a human search term such as "Detached Bungalow in a rural area" and it'll find the most suitable result based on the agent's description!