Project Portfolio
CloudPrune AI - FinOps SaaS Platform 🚀
Production B2B2B SaaS | AWS CDK + Lambda (FastAPI + Mangum) + ECS Fargate + Next.js 14 + async PostgreSQL + Claude API + Stripe + Turbo
Production B2B2B FinOps SaaS platform that analyzes AWS accounts and auto-generates Infrastructure-as-Code in 3 formats (CDK Python, Terraform HCL, CloudFormation YAML) for cost optimization, distributed via a partner-led GTM model.
Key Features: 18 resource analyzers covering 31+ optimization checks across compute, storage, networking, databases, commitments, extended support, and AI/ML; AI-powered IaC code generation using Claude API with business context enrichment via Claude Haiku; real-time anomaly detection engine (3-sigma against a trailing 30-day baseline, EventBridge cron, Slack/Teams alert delivery); AI cost tracking for Bedrock, OpenAI, and Anthropic spend; hybrid serverless architecture (Lambda + Mangum + ECS Fargate); 271-test suite with CI/CD via GitHub Actions OIDC; white-label PDF reports; tiered pricing (Free / Assist €4,900 / Operate €590/mo) with gainshare; GDPR-compliant architecture with DPA, ROPA, and Security Posture Document.
Validation: 4 active partners across Argentina, Spain, and Uruguay. First real partner scan detected USD 5,260/year in waste across a client AWS account. USPTO trademark registration in progress.
Visit Website
Company LinkedIn
Quira - Industrial IoT Platform 🚀
Production Multi-Tenant SaaS | AWS CDK (9 stacks) + ECS Fargate + ALB + WAF + Cognito + RDS PostgreSQL 15 + FastAPI + Next.js 14 + ESP32 (Arduino C++)
Multi-tenant industrial IoT platform for environmental monitoring (temperature, humidity), with real-time alerts, auditable PDF/CSV reports, and native Microsoft Teams integration via Power Automate. Built solo from architecture to production launch in under 4 months.
Key Features: 9 CDK stacks on AWS (ECS Fargate behind ALB with WAF, Route53 + ACM, RDS PostgreSQL 15, Cognito, SQS, 6 Lambdas, SNS); FastAPI backend with 47 endpoints and 103 tests; HMAC-SHA256 + HKDF device authentication with idempotent ingestion; alert engine with severity escalation; async PDF/CSV report generation via WeasyPrint; sensor-agnostic data model and driver registry (CO2, pressure, vibration on V2 roadmap); Next.js 14 admin panel with 25 pages and full ES/EN i18n.
Validation: End-to-end validated with real ESP32 + SHT31 hardware. First partner LOI signed in Colombia (aerospace warehouse pilot, ~30 devices). Production launch: April 19, 2026.
Visit Website
Multimodal Clinical Inference Platform
AWS CDK (Python, 6 stacks) + Step Functions + Bedrock (Claude) + Comprehend Medical + Textract + FHIR
Event-driven pipeline that turns a manual, per-disease clinical-variable extraction process into an auditable, disease-agnostic "factory" — from raw medical documents to ML-ready gold feature matrices and a cohort-selection API for clinical trials.
Architecture: S3 → EventBridge → Step Functions with 4 processing layers — multimodal normalize (OCR/text/HTML/structured), FHIR encoding to ICD-10 / RxNorm / SNOMED, LLM-based patient-journey canonicalization, and a gold feature matrix + cohort API.
Key decisions: deterministic where possible / LLM only to read prose (with a deterministic production fallback); model-as-configuration validated by a controlled 4-model benchmark; governance-as-code (CDK Aspects, OIDC CI/CD, KMS CMKs, DLQs); HIPAA/PHI compliance-by-design (Bedrock BAA, zero data retention, PHI redaction, surrogate patient keys). High-accuracy cohort extraction beating the deterministic baseline.
Built for: Argeniss Software (client under NDA)
Consumer-Finance Data Platform (AWS)
Terraform + Glue (PySpark) + AWS DMS + Redshift Spectrum + S3 + Step Functions + DynamoDB
Config-driven data platform replicating multiple SQL Server databases and MongoDB into an S3 data lake, exposed as a medallion architecture in Redshift — moving analytics off production databases and giving the business row-level change auditability it never had.
Key Features: medallion layers (RAW via Spectrum → SILVER views → GOLD materialized tables with DISTKEY/SORTKEY); DynamoDB-driven extraction catalog with three modes (full_load / incremental / SQL Server change-tracking); per-row SHA-256 diff emitting an INSERT/UPDATE/DELETE changelog in Parquet; SQL Server → Redshift query translation; least-privilege IaC with SSM + KMS.
Impact: heavy reporting cut from minutes to milliseconds (multi-thousand-x speedups); onboarding a new table went from a code change to a single config item; passed the client security review with zero open findings.
Built for: Argeniss Software (client under NDA)
Veterans-Benefits Conversational Form-Fill Assistant
AWS Bedrock AgentCore (multi-agent, Claude Sonnet) + WebSocket API Gateway + DynamoDB + React
LLM-powered chat that walks veterans through complex government benefits forms end-to-end — from automatic field extraction to a fully-filled, submission-ready PDF — with an admin surface where non-technical SMEs edit forms and delivery policy without touching code.
Key Features: separate Bedrock agents for entry recommendation, form completion, PDF preprocessing, and schema extraction; WebSocket chat with keep-alive heartbeat; DynamoDB single-table design; advanced prompt engineering (persistent cross-turn flags, one-shot resume directives). Owned a 48-hour pre-demo hardening cycle — root-caused a WebSocket session-loss chain and shipped atomic PRs with per-PR smoke tests and zero rollbacks.
Built for: Argeniss Software (client under NDA)
Labor Forecasting & Capacity Planning SaaS
FastAPI (Lambda) + React SPA + RDS Postgres (VPC) + Cognito RBAC + custom MCP server + Athena
Full-stack SaaS replacing an Excel-and-macros workbook (~470 projects, 5-year horizon) as the single source of truth for labor forecasting and hiring across a US mechanical contractor's offices and trades. Built solo, end-to-end.
Key Features: monthly/quarterly forecast entry with forward-fill cascade; automatic actuals pull from a legacy ERP over SOAP with a review + approval screen; capacity-vs-demand dashboard (utilisation, gap, required headcount per office × trade); natural-language scenario planning via a custom MCP server (what-if forecasts promoted into canonical estimates); Cognito RBAC with a custom invite flow. 8 CDK stacks, delivered iteratively across dozens of PRs.
Built for: Argeniss Software (client under NDA)
Moovelix — B2B Logistics Marketplace
React + TypeScript + Vite | FastAPI + Python | PostgreSQL + PostGIS | AWS ECS Fargate
Full-stack SaaS marketplace connecting cargo shippers with carriers in Argentina's livestock transport sector. Built end-to-end as sole developer.
Infrastructure: Dual-service ECS Fargate deployment (frontend + backend), AWS CDK for IaC, complete CI/CD with GitHub Actions, S3 for document storage.
Integrations: MercadoPago payments, Google Maps (Places, Geocoding, Distance Matrix), SendGrid transactional emails.
Key Features: 7 user roles with granular permission matrix, geospatial search with PostGIS (find loads near your location), "Return Trip" feature to reduce empty miles, subscription model with 60-day trial + paywall, complete document management with expiration tracking, logistics company module (fleet management, driver assignments, operator portal), 50+ REST endpoints with Pydantic validation.
Business Model: Freemium → $30/vehicle/month subscription
Status: Staging (~65% complete)
AWS CDK Project
CDK + Lambda + s3 + DinamoDB + React + CloudFront
This project implements a web application to manage NGO locations, allowing visualization, creation, updating, and deletion of locations on an interactive map.
The architecture uses core AWS services (DynamoDB, Lambda, API Gateway, S3, and CloudFront) with infrastructure as code through CDK, while the technical implementation consists of a Python backend with Lambda/DynamoDB and a React/TypeScript frontend using Vite and Leaflet for maps.
View on GitHub
AI-Powered Student Performance Advisor
AWS Bedrock (Claude) + Lambda + API Gateway + DynamoDB + S3
Intelligent analysis system that evaluates student performance data and generates personalized study recommendations using LLMs. The system uses a dynamic configuration architecture where prompts, endpoints, and output formats are stored in DynamoDB — enabling non-technical teams to modify AI behavior without code changes.
Key technical decisions: containerized Lambda functions, VPC peering for secure database access, incremental ETL with timestamp-based deduplication, and Parquet-optimized data lake with Glue/Athena for analytics.
Built for: Argeniss Software (client: confidential)
Document Intelligence Pipeline
AWS Bedrock (Claude 3) + Textract + Neptune + Step Functions + CDK
End-to-end document processing pipeline that extracts structured data from legal and medical documents using OCR (Textract), performs semantic analysis with LLMs (Bedrock), and builds knowledge graphs (Neptune) for cross-case analytics.
Architecture: dual graph system (mini + global Neptune clusters), automated pipeline orchestrated with Step Functions, PII controls with metadata-driven governance, full Infrastructure as Code with CDK Python.
Built for: Argeniss Software (client: confidential)
Dual Forecast System with SageMaker
SageMaker Canvas + Lambda (Docker) + S3 + CDK + GitHub Actions CI/CD
Machine learning forecast infrastructure with dual prediction models (short-term 180 days / long-term 12 months). S3 event-driven architecture triggers containerized Lambda functions that invoke SageMaker endpoints for real-time predictions.
Features: multi-environment deployment with CDK parameterization, automated model registry, S3-based input/output separation, full CI/CD with GitHub Actions.
Built for: Argeniss Software (client: confidential)
AWS Data Warehouse & QuickSight Dashboards
AWS CDK + Lambda + Glue + Athena + QuickSight + EventBridge + S3
Serverless ETL pipeline that extracts data from external APIs and SQL databases, transforms to Parquet format, and loads into an S3 data lake with Glue catalog. Incremental processing with DynamoDB-based metadata tracking and EventBridge scheduling.
Analytics layer with Athena queries and QuickSight dashboards combining historical and real-time data. Infrastructure fully defined as code with AWS CDK Python.
Built for: Argeniss Software (client: confidential)
HL7 and FHIR Healthcare Project
Python + HL7 + FHIR
HL7 to FHIR Converter: Streamline healthcare data interoperability. Transform HL7 ORU messages to FHIR resources with this two-step Python toolkit. Simplify lab result integration and enhance data exchange in modern healthcare systems.
View on GitHub
