Hyalen M. Caldeira

Senior AI Engineer
Agentic Systems · GenAI · RAG · MCP
- EST. 2026 .v1 -

.Portfolio

Some of the most recent projects I've selected.

.About Me

I'm a Senior AI Engineer with 15+ years of experience building production systems across Full-Stack Development, Cloud Infrastructure, and DevOps/SRE. Today I work as a Forward Deployed Engineer focused on multi-agent AI Platforms, GenAI, RAG pipelines, and MCP infrastructure - owning AI systems end-to-end from prompt and context engineering through deployment, evaluation, and observability.

Recent client work includes a reusable MCP blueprint (SSE, HTTP, STDIO transports) adopted by 8+ engineering teams that cut new AI project setup time by ~40%; a LangGraph + LangChain + MCP multi-agent code generation platform with JSON-configurable workflow routing that reduced feature delivery time by ~30%; and an AI-powered COBOL Modernization system that collapses weeks of manual analysis into days. I instrument production workflows with LangSmith and Langfuse for trace-level debugging, prompt regression testing, and quality monitoring.

Outside of client work, I build open-source AI tooling: a source-available platform for reverse-engineering enterprise COBOL applications by combining Compiler Theory, Graph-Based Program Analysis (AST, ASG, CFG, DFG, PDG), and selectively orchestrated LLM Workflows; DocWise, a RAG-powered documentation assistant with cascading retrieval (dense + sparse + rerank) over Chroma and Pinecone; and an iFlow XML Parser (SAP CAP + GenAI) that turns SAP Integration Suite workflows into structured documentation, user stories, and developer artifacts.

Beyond AI, I've built self-healing platforms, improved Kubernetes reliability, automated infrastructure with Terraform and Helm, and implemented CI/CD pipelines and observability stacks using Splunk, Grafana, Prometheus, and Dynatrace.

My background spans mission-critical Government Applications, high-traffic Ad Network APIs, E-commerce Platforms, and SAP BTP/CPI integration projects connecting enterprise systems. I've built full-stack applications with Python, Java, TypeScript, and Spring Boot, modernized legacy systems, and delivered scalable RESTful APIs across diverse business domains.

I work in Agile environments with a strong focus on Test-Driven Development, delivering production-grade solutions that teams can depend on.

I'm a builder at heart - I thrive on creating scalable systems, automating repetitive workflows, improving developer experience, and shipping software that solves real problems.

.How to contact me

For business contacts, collaborations, quotes or just to say "hello" write to:

hyalen@gmail.com

You can also find me on:

DocWise — RAG Documentation Assistant

Production RAG application that ingests documentation from any URL and serves context-aware conversational answers. Built on a protocol-based, swappable architecture so embedding strategies, chunkers, and vector stores are configurable without code changes.

Built on LangChain, GPT-4o-mini, and OpenAI embeddings;
Dual vector store: Chroma for development and Pinecone for production (1536d cosine);
TavilyExtract-first ingestion pipeline with automatic BeautifulSoup fallback for resilient web scraping;
Implements cascading retrieval (dense + sparse + rerank) and conversational memory for multi-turn context;
Full LangSmith tracing for prompt debugging and quality monitoring;
Streamlit UI for interactive querying.

Project: DocWise
Client: Independent Project
Date: 2025-2026
Role: AI Engineer & Developer

iFlow XML Parser — SAP CAP + GenAI

SAP CAP application (TypeScript + Node.js + CDS, deployed to SAP BTP via MTA) that parses SAP Integration Suite iFlow XML into a doubly-linked-list step graph, preserving exact execution order, then uses generative AI to produce structured documentation, user stories, and developer artifacts.

Combines deterministic XML parsing with OpenAI-driven synthesis to eliminate hallucination on control flow;
Doubly-linked-list representation preserves exact iFlow execution order for accurate downstream documentation;
Exposes both REST and SOAP (WSDL-driven) interfaces;
Cuts manual integration analysis from hours to minutes per iFlow.

Project: iFlow GenAI
Client: Independent Project
Date: 2025-2026
Role: AI Engineer & Developer

AI-Driven COBOL Reverse Engineering Platform

Architected and implemented an AI-Powered platform for reverse engineering legacy COBOL systems into structured, analyzable representations, bridging compiler theory with modern agent-based AI Architectures. This project is developed independently, outside of my employment and client responsibilities.

Built compiler-style analysis pipelines (AST, ASG, CFG, DFG, PDG) to extract deterministic program structure and semantics;
Designed graph-based representations for cross-program dependency analysis and system-level understanding;
Implemented agent-driven workflows using LangGraph to orchestrate multi-step analysis and reasoning pipelines;
Integrated LangChain for controlled LLM interactions, enabling selective AI usage while preserving cost efficiency and determinism;
Designed and implemented an MCP-based modular tooling architecture, enabling scalable tool execution, and future extensibility;
Added enterprise-grade observability and execution metadata (timing, runs, errors) to support debugging and operational insights.

The platform combines Compiler Theory, Graph-Based Program Analysis, and selectively orchestrated LLM Workflows, and is exposed via Model Context Protocol (MCP) servers consumable by MCP-compatible clients such as Cursor and Claude.

MCP Endpoints:

MCP Server (HTTP transport): https://mcp.hyalen.us:8003/mcp
MCP Server (SSE transport): https://mcp.hyalen.us:8001/sse
Health: https://mcp.hyalen.us/health
Prometheus Metrics: https://mcp.hyalen.us/metrics

GitHub Repository:

https://github.com/legacy-ai-labs/mcp-server-language-converter

Project: COBOL AI Platform
Client: Independent Project
Date: 2025-2026
Role: AI Architect & Developer

MCP Infra AWS — Terragrunt IaC

Production-grade Terragrunt IaC deploying the MCP Server to AWS (EC2, Elastic IP, IAM, Route 53) with S3-backed remote state and DynamoDB locking. Companion infrastructure to the open-source MCP Server Language Converter project.

Hierarchical configuration (account → environment → region → service) enables clean multi-environment expansion;
Makefile lifecycle automation for predictable, repeatable deploys;
EC2 bootstrap runs Docker Compose (PostgreSQL, MCP server, NGINX) with optional Let's Encrypt SSL;
Remote state and locking ensure safe concurrent operations across environments.

Project: MCP Infra AWS
Client: Independent Project
Date: 2025-2026
Role: Cloud & Infrastructure Engineer

Hyalen M. Caldeira

.Portfolio

.About Me

.Certificates

.How to contact me

hyalen@gmail.com