Projects

Brought the teachings from IBeeLearning into a teacher-centric tool for lesson planning. Ditched the langchain overhead, built the flows from scratch and life was suddenly easy once again. Then got to the part where markdown,latex, special characters, etc needed to work across views, edit, pdf, clipboard and life returned to the expected complexity.

A humble proof of concept trying to bridge the affordability gap in education using LLM tools. By automating the most straightforward parts of the teacher-student work process, we create opportunities in two directions: 1) students who cannot afford the private teaching can self-serve and get similar results at a fraction of the cost; 2) teachers can streamline their processes, creating space for more meaningful interaction with students.NEXTJS. FULLSTACK. AGENTIC WORKFLOWS. Got more serious with AI and nextjs as one should it the year of 2024. Constant battle between even improving AI tools and AI ecosystem that somehow could not catch up.

Choosing snowboard circa 2023 was not a great experience. The advent of semi-capable LLMs meant that this could be addressed in a pretty straightforward way. LLM-powered scrapers, custom RAGs, first steps in server components (with all the Nextjs jank at this time) - it was a fun project. Much of what it does now can be produced by smarter AI. This, combined with the peculiar nature of monetizing through referrals, moved it to the the furthest of back burners

The culmination of my first startup attempt. Started as a little internal little tool to replace our airtable bootleg CRM, it has grown into a full fledged platform offering a CRM for football agents combined with all the models and services we have previously developed. As the purpose and complexity changed, we migrated from Mongo to Postres and developed organization support and all that jazz. Since my departure from the company, my colleagues have brought it to a whole new level and are serving some of the most big baller football agencies out there.

The backbone of data collection and generation for ensk.ai, Arion is a pipeline that processes data from the company's web crawlers, APIs of data vendors and our own batch model scoring jobs. I was punching way beyond my weight class at the time and I learned a whole bunch of bad developer practices in return. Some of the fanciest and best organized code I have written (this was before the rise of the vibes) - and most of it still running smooth as butter 5 years later

Manufacturing Order Optimization Management System for a good friend of mine. We built an app that adds, tracks orders&designs for printing yogurt lids, butter labels, and others, and performs fairly complex optimization (i.e. python brute force permutations until the docker blows up) of the printing process, given constraints of urgency, time, material, colors, machine/employees availability. Did it sveltekit+supabase and felt pretty hip about it

Full rewrite of an entire existing beckend from PHP monolith to python microservices for GroupM. Rolled out custom auth and authorization, gateway, and replaced over 20 CRUD and analytics microservices. Wrote most of it in a bamboo hut in the Philippines, when my body was too battered to kitesurf.

My humble beginnings in the wonderful world of data science. All sorts of regression analysis, ML models, forecasts, dashboards, ppts for industry giants like Coca Cola, IKEA, Adidas, Liberty Global, Balenciaga. Many egregious offenses against data science were committed, sometimes actual science was miraculously accomplished, but no matter the case, client satisfaction was unmatched. Taught me that I was not cut for the consulting excellence.