jose tavera dev sheet
I’m Jose Tavera. At Quantil I build LLM applications, forecasting models and recommendation systems for enterprise clients across the US and LatAm. On the side I ship my own products, solo.
sec 02 · specimen
1-bit · 8×8 bayer
I got here through economics rather than a CS degree, which mostly means I learned to ask what a number is for before asking how to compute it. These days I build the whole thing — the model, the pipeline that feeds it, and the product it sits behind.
commits · jl-tavera
synthetic · run `npm run github`
audio · on repeat
sec 03 · catalogue
Client engagements and the products I ship solo under Mayeutik. Each card names its own context.
stn 01 · quantil — client engagement
Estimated the causal effect of drilling practices and mud parameters on well performance across 17 wells, then separated the levers a driller controls from the geology they cannot.
stn 02 rec 2026 · orrery — home health compliance
Compiled the Medicare home-health pre-claim review requirement set into 108 machine-readable rules, each carrying its own regulatory quote, citation and pass/fail contract — so an agency can see what is missing before it submits, not after it is denied.
stn 03 · quantil — client engagement
Built the model that turns a production plan into the electricity it will need — across ten operating units, three planning horizons, and six years of daily data that had never been assembled in one place.
stn 04 rec 2024 · universidad de los andes — coursework
Built a 120-breed dog classifier the whole way out — scraper, two transfer-learning architectures, a FastAPI service, a Cloud Function and a Streamlit front end — then found the pretrained backbone had already solved it before the first epoch finished.
sec 04 · index
Filed by what the work is, not by what is installed. Open one for the tools it uses and an example of what the code looks like.
Two-tower retrieval and implicit ALS for Panama's largest logistics company, deployed on Snowflake to replace a legacy rules-based system. The tower pattern is the useful part: users and items each get their own encoder, so serving becomes a nearest-neighbour lookup instead of scoring every pair.
plate 03.1
listing · python
embedding_dimension = 32 user_model = tf.keras.Sequential([ tf.keras.layers.StringLookup( vocabulary=unique_user_ids, mask_token=None), # +1 accounts for the unknown token. tf.keras.layers.Embedding( len(unique_user_ids) + 1, embedding_dimension),]) movie_model = tf.keras.Sequential([ tf.keras.layers.StringLookup( vocabulary=unique_movie_titles, mask_token=None), tf.keras.layers.Embedding( len(unique_movie_titles) + 1, embedding_dimension),]) metrics = tfrs.metrics.FactorizedTopK( candidates=movies.batch(128).map(movie_model))task = tfrs.tasks.Retrieval(metrics=metrics)The two encoders are the architecture. Everything else in a retrieval model is plumbing around them.
source
Canonical example, not client code. TensorFlow Recommenders, basic_retrieval tutorial · tensorflow.org
instruments
sec 05 · service record
ML and AI engineer who builds products end to end. At Quantil I build LLM applications, forecasting models and recommendation systems for enterprise clients across the US and LatAm.
On the side I keep shipping independent AI products to stay sharp — the point is to own the whole loop: pick the problem, scope an MVP, ship it, see what breaks, iterate.
Senior ML Researcher
Bogotá, Colombia · HybridProduction AI for enterprise clients across the US and LatAm — LLM applications, forecasting models, recommendation systems and large data pipelines.
AI Engineer (Independent)
Bogotá, Colombia · RemoteIndependent AI products shipped solo end to end: pick the problem, scope an MVP, ship, iterate. Same code quality bar as production work.
Co-Founder and Technical Lead
Bogotá, Colombia / St. Gallen, SwitzerlandCo-founded a practice management SaaS for therapists and psychologists across LatAm. The platform is still operating today. First and only product hire; owned product, engineering and go-to-market.
START Fellow
University of St. Gallen, SwitzerlandSelected for the START Fellowship Accelerator at the University of St. Gallen — a 4-month residential program taking around 30 founders from LatAm and Africa each year through a multi-round selection. Went through it as technical co-founder of Psiris.
Teaching and Research Assistant
Bogotá, ColombiaFour appointments across the economics and mathematics faculties, plus the introductory programming support centre.
sec 06 · annotations
distinctions
AI Hackathon Winner — Best WhatsApp Agent
Colombia Tech Week
Production-ready conversational agent built in 48 hours.
International Fellow
START Fellowship, University of St. Gallen · 2022 — 2023
Outstanding Delegate
Harvard Model United Nations for Latin America
Silver Medalist
American Junior Science Olympiad · 2016
Barranquilla, Colombia. Member of the Colombian team at the International Junior Science Olympiad.
education
Bachelor's Degree, Economics
Universidad de los Andes, Bogotá · 2018 — 2024
Member of ONUAndes and the Analytics Club.
Minor in Law
Universidad de los Andes, Bogotá · 2018 — 2020
Exchange semester, Economics
Universität St. Gallen (HSG) · 2023
High School Diploma
Colegio San Carlos, Bogotá · 2018
certifications
Startup School
Y Combinator
languages
Spanish
Native
English
C1 · IELTS
Runs client-facing meetings in both, for clients across the US and LatAm.
sec 07 · reference
Open to ML and AI engineering work — production models, LLM applications, and the pipelines that keep them fed. The fastest way to reach me is email.