sebastian@harmony ~/retail-vision %
detect --products --country=co products detected 142
brands matched 37
shelf share 84%
sebastian@portfolio: ~/ai-agents
sebastian@portfolio ~/ai-agents % whoami
Sebastian Torres Ortiz
AI Agent Systems Engineer | Machine Learning Engineer | Automated Solutions Engineer
sebastian@portfolio ~/ai-agents % cat ./bio.txt
300+
brands served
5M+
predictions served
↑47%
efficiency gain
<12ms
inference latency
exit 0 · portfolio v2.0.0 · astro + react + typescript · retail ai
contact --sebastian Available for AI agents, ML systems, and backend automation work.
projects / active deployment
Retail Intelligence In Production
Harmony · ColombiaLeading the machine learning work for Harmony, building product recognition systems that turn retail shelf images into detection, coverage, and brand execution analytics.
agent map financial research automation
Router
Classifies user objective and dispatches the research flow.
User Profile
Reads capital, horizon, risk tolerance, and strategy constraints.
Market Intel
Combines price action, sentiment, macro, and structural signals.
Strategy Builder
Generates entries, invalidation, sizing, and trade scenarios.
Risk Engine
Stress-tests drawdown, exposure, leverage, and stop placement.
Analyst Report
Returns thesis, plan, risk notes, and market analysis.
prompt in
“Build a BTC/ETH strategy for my risk profile, include invalidation and market context.”
strategy state validated
risk mode controlled
analysis loop live
experience / skills
Sebastian builds AI agent systems with full-stack engineering depth.
This portfolio is about the systems I build: production AI agents, machine learning workflows, backend services, cloud infrastructure, databases, and frontend interfaces that make automation usable.
Machine Learning Head at HarmonyQuant ML anomaly detection · Reals Lab PropTech modelsProduction systems across backend, cloud, ML, and UI
Sebastian Torres Ortiz 0+ years of experience
Software developer focused on AI agents, ML products, and production-grade automation.
01 market + business data
02 agent planning
03 backend services
04 databases
05 cloud deployment
06 frontend interfaces
07 linux
sebastian@agent-lab ~/build %
ship --agents --backend --cloud --frontend Agentic Systems
Backend Engineering
Data & ML
Cloud & Infra
Frontend Product
experience / real estate ml
Automated valuation models for Colombian apartment markets.
Built an XGBoost AVM that predicts price-per-m² from listing attributes, cadastral references, and locality signals — with explicit data-quality audits before training.
Ingest platform listings + cadastral
Audit 420 garage · 216 stratum fixes
Features locality · geo · cadastral/m²
Train XGBoost · price/m² target
Valuate apartment AVM inference
top feature importances
Stratum38.5%
Antiquity23.8%
Cadastral per m²21.3%
Garage13.4%
Area (m²)8.1%
Rooms5.3%
data audit 420 garage outliers capped · 216 stratum values normalized
train / test split 2,984 / 747
test rmse $374M COP
target price/m²
experience / listing quality ml
Automated audit of property listings on Colombia’s top platforms.
Built segment-specific XGBoost models that flag anomalous listings scraped from Finca Raíz and Metro Cuadrado — detecting pricing, attribute, and cadastral inconsistencies before they reach downstream systems.
Scrape Finca Raíz · Metro Cuadrado
Label 2,420 anomalies flagged
Segment city × property type
Train XGBoost per segment
Score anomaly probability
hold-out performance by segment
ROC 93.3% F1 74.4% Acc 85.1%
ROC 92.9% F1 85.8% Acc 89.2%
ROC 89.4% F1 84% Acc 83.8%
ROC 97.8% F1 83.3% Acc 96.8%
ROC 95% F1 57.1% Acc 92.7%
ROC 92.2% F1 75.6% Acc 85.3%
label distribution 5,848 passed · 2,420 rejected · threshold 50%
features m² · geo · cadastral
avg f1 76.7%
run id 20260525
experience / quant ml · restricted
Quant machine learning for market anomaly detection.
Built production quant ML pipelines that flag structural anomalies across crypto, S&P 500, and gold — trained on 10+ years of multi-asset history. Live performance metrics remain confidential.
Ingest 10+ yr OHLCV · macro feeds
Features regime · vol · momentum
Train quant ML detectors
Score anomaly probability
Deploy production · classified
monitored universe
production metrics RESTRICTED
live detection precision ██.█%
production alert rate ███ / day
inference latency < ██ ms
portfolio symbols live ██ active
Live trading analytics, alert thresholds, and model scores are withheld — client-confidential production data.
history window 10+ years
symbol count 10+ assets
objective anomaly detect
experience / harmony geospatial
Retail census simulation for Bogotá’s store network.
At Harmony, I worked on data systems that collect more than 40k stores in Bogotá and transform them into simulation maps, density layers, microzones, and 1,000+ isochrones for a realistic census of the city.
retail density high activity corridor
1k+ isochrones microzone simulation
stores mapped 40k+
microzones 1k+
city census bogotá
experience / blockchain gaming
Unreal Destiny RPG with a decentralized game economy.
I worked as a smart contract developer on Unreal Destiny, an RPG where blockchain was used to keep the game economy transparent, auditable, and decentralized across multiple networks.
Player Action quest / trade / reward
Smart Contract validates economy rules
Blockchain transparent settlement
contact / next collaboration
Want to know more about Sebastian?
Send a short note about the system you want to build. The fastest path is email, and the form below prepares a clear message you can send immediately.
Best easy setup for a static portfolio: use this mail draft now, or connect the same UI to Formspree, Netlify Forms, or Resend when you want inbox delivery without opening an email client.