2026
BlackPrint Technologies
mireporte.ai, Read the case
Turns a point on the map into a frozen 25-section market study across 11 metro areas, kept honest by 5,043 passing tests.
- 11 Mexican Metro Areas Coverage
- 5,043 Passing Tests Test Suite
Now: Lead AI Engineer at BlackPrint Technologies
Gairo Peralta I build AI that reads places and language.
LLM agents, GeoAI platforms and production ML pipelines that turn maps, documents and conversations into answers your team can check.
AI/ML Engineer LLM systems & agents
Where I've worked and studied
- BlackPrint Technologies
- Law Offices of Manuel Solis
- UT San Antonio
- UC Berkeley
- Tecnológico de Monterrey
- STRTGY
- PRODENSA
Measured outcomes
Every number, with its project.
- urban blocks with real-time inference LLM GeoAI Platform · BlackPrint Technologies
- 2.5M+
- merchants scored for payment-acceptance propensity Acceptance Radar · BlackPrint Technologies
- ~5.5M
- faster legal drafting: ~2 h → ~15 min LLM agents · Law Offices of Manuel Solis
- ~8×
- medical QA accuracy, reaching 54.69% Multi-agent QA on PrimeKG · UT San Antonio
- +24.2pp
Selected work
From city blocks to case files.
Five projects from 2025–2026: territory intelligence, field-sales routing, payments data at national scale, legal AI and medical research.
2026
BlackPrint Technologies
Pharma Field Sales, Read the case
Driving-aware daily routes, GPS-verified pharmacy visits with photo evidence and supervisor review for a five-role sales team.
- 3,100+ Pharmacies Pilot Universe
- 260+ Endpoints API Surface
2026
BlackPrint Technologies
Acceptance Radar, Read the case
Scores payment-acceptance propensity across ~5.5M merchants; a 45 GB extract now ships as a 70 MB Parquet file.
- ~5.5M (National) Merchants Extracted
- 594,068 CDMX POIs Launch Scope
2025
Law Offices of Manuel Solis
DeclarationLetterOnline, Read the case
Turns questionnaires into court-formatted declaration letters in under 60 seconds, with a local audit trail.
- <60s Processing Time
2025
UT San Antonio (Volunteer)
Medical QA Multi-Agent System, Read the case
Four agents reason over PrimeKG (4M+ nodes) in Neo4j and lift medical QA accuracy 24.24 points, to 54.69%.
- 54.69% Accuracy
- +24.24pp Lift
What I build
From the data warehouse to the map on screen.
Most projects need all four: mireporte.ai turns a point on a map into a frozen 25-section market study built on ten data layers.
GeoAI & spatial intelligence
Models and maps for where-questions: site selection, competitive attraction, POI matching and risk indices, from H3 cells to census blocks.
- MapLibre GL
- H3
- GeoPandas
- Turf.js
LLM systems & agents
Multi-agent orchestration, RAG and document pipelines: verified SQL, audit trails, role-based access and a handoff to human operators.
- LangGraph
- RAG
- Pinecone
- Neo4j
Production ML & forecasting
Models that run behind an API, not in a notebook: demand forecasting with +42% accuracy and 30% lower inventory cost, and a site-sales model whose leave-one-out error fell from 106% to 22%.
- PyTorch
- XGBoost
- FastAPI
- MLflow
Data platforms & pipelines
Warehouses and ETL built to hold up: BigQuery with cost gates, Redshift with a CSV fallback, anonymized exports under Mexico's LFPDPPP.
- BigQuery
- Redshift
- Terraform
- pandas
Path
From Berkeley data science to AI in production.
-
2022
BA in Data Science, UC Berkeley
-
2023
AI/ML Engineer at STRTGY: GeoAI and RAG
-
2024
OCR + Llama 3.3 extraction for PRODENSA
-
2025
Back to BlackPrint; legal AI lead; UTSA research
-
2026
mireporte.ai and GeoAI products at BlackPrint