FM Fernando Muerza ML Engineer
Contact
Navigation
01About 02Skills 03Projects 04ML Valuator 05Track record 06Notes
Language
Contact →
Puebla · CDMX, MX
Machine Learning Engineer — Production ML · Real Estate · AI

Fernando
Muerza

I build ML for pricing, valuation and attribution — and ship it to production as APIs, not notebooks.
Python, SQL, Docker and REST APIs with scheduled retraining across real estate and performance marketing — now extended with LLMs, RAG and agents on AWS. Measured, never guessed.
Open to work · Puebla & CDMX · hybrid or remote (CT/ET overlap)
AWS Certified AI Practitioner 2026 ↗
View projects →
Contact me
R² · valuation
0.904
Cost / lead
−70.76%
Valuation · was 4 h
39s
Valuation report Sample
House · Lomas de Angelópolis, Puebla
Estimated value
$8,043,200
80% INTERVAL±10%
$7.24M$8.85M
R²
0.904
Retraining
7 days
Inference
39 s
Model
LightGBM
History+18% / 12m
In production · scikit-learn · Django / Docker
R² · real-estate valuation
0.904
Cost per lead · attribution
−70.76%
Valuation in production
39s
before · 4 h
Comparable search · KNN
120×
before · 8 h → 4 min
§ 01 — About
Fernando Muerza, Machine Learning Engineer
Fernando MuerzaPuebla · CDMX
SpanishNative
EnglishB1 → B2, improving

Machine Learning Engineer focused on financial outcomes, end to end.

I build machine-learning solutions aimed at financial results: revenue, margin, acquisition cost, retention and ROI. I work end to end — from the business problem and clean, representative data to a model deployed as an API in production. I close a project on the return it generates, not on model metrics alone.
In production: R² = 0.904 in real-estate valuation and a −70.76% cut in cost per lead. Before modeling, I solved real problems in marketing, sales, operations and automation, so I read the business in its own language: unit economics, efficiency, return per channel. AWS Certified AI Practitioner, TripleTen Data Science certificate, and formal training in valuation (Damodaran · NYU Stern; IBERO Puebla).
Domain
Real Estate
valuation & classification
Domain
Applied AI
RAG, agents & AWS
Domain
Performance
multi-channel attribution
§ 02 — Skills

End-to-end solutions, delivered to production.

01
Predictive ML
—Data cleaning, EDA and validation
—Feature engineering and SQL (window functions)
—Gradient boosting: LightGBM, XGBoost, CatBoost
—Time series, classification and regression
02
LLMs & AI engineering
—RAG pipelines and LLM evaluation
—Agents with LangChain / LangGraph, tools and MCP
—Deep learning: CNNs (ResNet50), BERT
—AWS · Certified AI Practitioner
03
Production & automation
—Model-serving APIs (Django, FastAPI)
—Docker and VPS deployment, scheduled retraining
—Automation: n8n, Make, webhooks, REST APIs
—Data-quality controls and leakage checks
Python Pandas scikit-learn LightGBM XGBoost PyTorch Keras LangChain AWS FastAPI Django Docker PostgreSQL n8n Make Git
§ 03 — Projects

Featured projects

github.com/muerza ↗
In productionInmuebles Anzuz · 2025 — present

Real-estate valuation

End-to-end valuation models: multi-source data preparation, feature engineering, data-quality controls and weekly retraining. Deployed as a Django API on Docker/VPS with 80% confidence intervals — plus a KNN similar-property search that went from 8 h to 4 min.
scikit-learnLightGBMDjangoDockerVPS
Try the ML Valuator →
R²0.904
Inference39 s
Before4 h
Retraining7 days
ConstructionIn progress↗

Cost-forecasting AI agent

Holt-Winters projects raw-material prices, regressions price the equipment, a FastAPI service exposes the data and a LangChain agent chains it all from a plain-language question.

LangChainFastAPIHolt-Winters
TravelIn progress↗

Tool-using AI agent

One Claude + LangChain agent that combines its own Python tools, Claude’s server-side web search and a remote MCP server to search, price and book hotels.

ClaudeLangChainMCP
Interconnect↗

Telecom churn prediction

Flags customers likely to cancel so retention spend goes to the highest-risk segments. Upsampling for class imbalance and an XGBoost + CatBoost + Random Forest ensemble.

AUC-ROC 0.847Ensemble
Good Seed↗

Age estimation from faces

ResNet50 fine-tuned on 7,591 faces (Keras 3 on a PyTorch backend) to support age checks at checkout, beating the ≤ 8-year error target.

MAE 5.57ResNet50
Sweet Lift↗

Hourly demand forecasting

Airport taxi orders forecast an hour ahead with LightGBM, 24 hourly lags and leak-free rolling means — 17% under the ≤ 48 RMSE target.

RMSE 39.7Time series
OilyGiant↗

Oil well selection under risk

Picks the region with the best expected profit: per-region regression plus 1,000-sample bootstrapping, keeping the probability of loss under 2.5%.

BootstrappingRisk analysis
§ 04 — Live demo

ML Valuator — Lomas de Angelópolis

Price estimated with LightGBM trained on real market data from Puebla. Adjust the parameters and get a valuation with its confidence interval.
LightGBM · live
Property parameters
Operation
Estimated price House · Lomas 2
$8,043,200
CONFIDENCE INTERVAL · 80%±10%
$7,239,000estimate$8,848,000
Top factors
Built m² $5.7M · 65%
Land m² $1.4M · 16%
Zone (Lomas 2) $731k · 8%
Bedrooms and bathrooms $528k · 6%
Base / location $385k · 4%
LightGBM model trained on Puebla market data. Illustrative estimate — not an official appraisal.
§ 05 — Track record

Experience & education

01
Nov 2023 — present Experience

Data Scientist · ML in Production

— Inmuebles Anzuz

Real-estate valuation ML models, taken end to end into production:

R² 0.904
Valuation model with LightGBM and scikit-learn; feature encoding of numeric and categorical variables from multiple sources
−99.7%
Valuation time (4 h → 39 s), exposing the model as a REST API in Django on Docker / VPS
120×
Similar-property search with KNN (8 h → 4 min), speeding up responses to clients
7 days
ETL pipeline with scheduled retraining to hold performance as the market shifts
Data
Data-quality controls and percentile outlier filters to stabilize training
02
Sep 2026 Certification

AWS Certified AI Practitioner

— Amazon Web Services

AI, ML and generative-AI concepts on AWS: choosing the right service for each use case, responsible AI, pricing and the shared-responsibility security model. Applied to: the RAG valuation system I’m building on AWS.

View credential ↗
03
Aug 2026 Education

AI Engineering with LangChain

— LangChain · LangGraph

Prompt engineering, evaluation, RAG, tool use and agents with LangGraph — the full map of how an LLM system fits together. Applied to: my first LangChain projects, a cost-forecasting agent and a tool-using hotel assistant.

04
2025 — ongoing Education

System Design Mastery

— ByteMonk Academy

System design focused on scalable architectures (load balancing, consistency) and hands-on case practice. Applied to: model APIs and services that are more scalable, reliable and production-ready.

05
Mar 2025 — Mar 2026 Education

Data Science Certificate

— TripleTen

Completed. Python, SQL and machine learning through deep learning and NLP, with a portfolio of end-to-end projects on GitHub: churn, forecasting, computer vision, NLP and risk.

06
May 2025 Certification

Make Advanced

— Make Academy

Robust automation: advanced webhooks, API calls with HTTP modules, pagination and data stores. Applied to: orchestrating ETL/MLOps integrations (ingestion, retraining, validation and alerting).

View credential ↗
07
Feb 2021 — Oct 2023 Experience

Commercial Operations Analyst

— Inmuebles Anzuz

Marketing analytics and funnel automation on campaign and CRM data:

−70.76%
Cost per lead on Meta Ads with server-side tracking (CAPI · Stape · GTM): $182.62 → $53.40 MXN
+72.32%
Funnel traceability with URL-parameter attribution in Zoho CRM
432×
Sales response time (36 h → 5 min) with lead capture and routing through APIs in n8n
13.65%
Lost conversion events recovered with server-side Meta CAPI (925 of 6,775 in 90 days)
Make
Contract generation with webhooks over CRM data, cutting errors and rework
08
2023 Education

Real-estate management

— IBERO Puebla

Formal training in real-estate management and appraisal — the domain foundation behind the pricing and valuation models.

09
2018 — 2022 Education

BBA, Business Administration

— UPAEP

Strategic and diagnostic foundation, with quantitative coursework (statistics, business mathematics). Applied to: turning technical findings into actionable decisions and communicating impact to non-technical stakeholders.

10
Aug 2018 — Jan 2021 Experience

Operations & Data Analyst

— Baoba

Automation of reporting and commercial processes at a panel and veneer manufacturer:

−5 h/wk
KPI reports and dashboards automated in Excel — (~260 h/year) of manual work removed
−85%
Time per quote (10 → 1.5 min) with a rule-based quoting system
−97%
Weekly process errors (35 → 1) with the same system
11
2019 — 2020 Education

Valuation & Corporate Finance

— Aswath Damodaran · NYU Stern

Valuation and corporate finance with a practical focus. Applied to: translating business and finance variables into defensible features and assumptions in proptech and fintech models.

§ 06 — Notes

Latest notes

Short updates from LinkedIn on what I’m building and learning — AWS, RAG and AI agents.
All posts on LinkedIn ↗
Certification Sep 2026

I’m now AWS Certified AI Practitioner

Verified by AWS: AI, ML and generative AI on AWS — the foundation for the RAG valuation system I’m building.

LinkedIn Read on LinkedIn →
AI engineering Aug 2026

Finished AI Engineering with LangChain

Prompt engineering, evaluation, RAG, tools and LangGraph agents: separately they are tools, together they are an ecosystem.

LinkedIn Read on LinkedIn →
AWS Summit Aug 2026

AWS Summit Mexico City, with a real problem in hand

I went with concrete questions on architecture, cost and deployment for a real-estate valuation system with RAG on AWS.

LinkedIn Read on LinkedIn →
§ 07 — Contact

Let's work
together

Have an ML Engineer opening, or a pricing, valuation or AI project in mind? Send me a message and I’ll get back to you.
Open to work · hybrid or remotePuebla · CDMX · CT/ET
Leave your details