Christian Callahan / Independent practice

Open to roles & project work

Data
engineer.

I design and build SQL and Python pipelines, data models, and reporting systems. I make the logic explicit, reconcile the results, and document the handoff.

See my work Let’s talk

Inside NY SPARCS

NY SPARCS starts with public, de-identified hospital discharge files. Field names change between annual releases.

For organizations big and small.
Across industries.

SQL / Python / Data pipelines

Selected work

Projects

Independent projects with public and synthetic data.
Try the apps or see how they work.

Graph and vector retrieval

Multi-Hop GraphRAG

Follow evidence across documents.

I combined graph traversal and vector search to connect evidence across documents. Compare retrieval paths and inspect the passages behind each answer in the interactive demo.

Pipeline map

SourceSynthetic documents
ConnectEntities + relationships
RetrieveKuzu + LanceDB
InspectPaths + source passages

View repository Read the approach

Multi-Hop GraphRAG demo showing retrieved evidence and graph traversal across synthetic documents — Project screenshot · synthetic document retrieval demo

Scope & evidence

Source: An independent project with 11 synthetic documents. The demo benchmark contains three questions.
Data flow: Document passages feed a Kuzu relationship graph and a LanceDB vector index. The app exposes retrieval paths and supporting passages for inspection.
Default: The default demo uses hash embeddings and heuristic answer extraction. Its answers are not generated by a language model.
Limits: The small synthetic benchmark illustrates the retrieval approach. It does not establish general accuracy or production performance.

Retrieval authorization

Permission-aware RAG

See how access rules change retrieval.

I built a retrieval demo that filters records before ranking them. Change the tenant, role, and clearance settings to see which synthetic records become eligible for retrieval.

Pipeline map

SourceSynthetic hospital records
FilterTenant + role + clearance
RankLexical cosine similarity
InspectAllowed source passages

View repository Read the approach

Permission-aware RAG demo showing access settings and retrieval over synthetic hospital records — Project screenshot · simulated access rules on synthetic data

Scope & evidence

Source: An independent demonstration using synthetic hospital records, with no patient data.
Data flow: Python filters records by tenant, role, and clearance before ranking the eligible passages with lexical cosine similarity.
Backend: The repository also contains a separate FastAPI backend using PostgreSQL row-level security, pgvector, and tenant audit chains. That backend is not running in this demo.
Limits: Streamlit settings simulate access rules. The demo does not authenticate users or enforce database row-level security, and it is not a production security assessment.

More projects

Other builds

Healthcare data engineering

NY SPARCS

I used Python and DuckDB to normalize New York’s annual discharge files and build Parquet reporting marts. The app explores sample data, not statewide totals.

View repository

Operational data modeling

ED Operations

I modeled Scotland’s emergency-department data by site and month in Python and DuckDB, then compared their results. The forecast’s uncertainty interval includes zero improvement.

View repository

Customer analytics

Ecommerce Retention & Growth

I built subscription features using data available before each prediction cutoff. Includes a real-data sample and a separate synthetic demo.

View repository

Financial modeling

AutoModeler

Financial data to Excel models with linked formulas. Stops when a balance-sheet adjustment exceeds the configured limit.

View repository

Text classification

Ticket Intel

TF-IDF and Naive Bayes for ticket classification. The app uses synthetic examples; code and evaluation are on GitHub.

View repository

Model evaluation

SignalForge

I compared churn models on IBM Telco data and documented how I evaluated them.

Read the case study

My approach

How I work

I build pipelines I can explain: where the data came from, how it changed, and how I checked the result.

Source
Understand.
I start with where it comes from, what each row means, what’s missing, and how often it changes.
SPARCS: annual schema changes
Model
Build.
I define the output before building the pipeline. Then I keep the extraction, staging, and transformation steps easy to follow.
ED Operations: the site-month model
Checks
Check.
I compare keys, counts, and totals against the source. If something doesn’t line up, the check should fail where I can see it.
ED Operations: SQL reconciliation tests
Hand-off
Document.
I document the sources, field definitions, refresh steps, and limits so someone else can pick it up without guessing.
SPARCS: source, build, and demo instructions

Background

Experience

My background is in healthcare reporting and business intelligence. That’s where I’ve worked on turning operational data into reports people can use.

Independent practice

CGC Labs

Founder / Data engineering

I started CGC Labs for my independent data engineering work. I focus on getting data out of legacy systems, cleaning it up, checking it, and documenting how it works.

Professional experience / McCook, Nebraska

Community Hospital

Business Intelligence Analyst

I built Tableau dashboards and automated recurring reports. I also created SQL staging for an electronic medical record migration and validation checks for billing data.

I keep this work high-level here. Employer code, internal systems, and patient data stay private.

Beyond this portfolio

Elsewhere.

Independent practice

CGC Labs

My independent data engineering work and consulting.

cgclabs.org

Product in progress

MeasureMap

I’m building a registry for metric definitions, owners, formulas, and source references.

measuremap.cgclabs.org

Contact

Let’s talk.

Hiring a data engineer or need help with a project?
Tell me what you’re working on.

Email mechristian.g.callahan@gmail.com Find me on GitHub

Christian Callahan / Independent practice

Open to roles & project work

Data
engineer.

I design and build SQL and Python pipelines, data models, and reporting systems. I make the logic explicit, reconcile the results, and document the handoff.

See my work Let’s talk

Inside NY SPARCS

NY SPARCS starts with public, de-identified hospital discharge files. Field names change between annual releases.

For organizations big and small.
Across industries.

SQL / Python / Data pipelines

Selected work

Projects

Independent projects with public and synthetic data.
Try the apps or see how they work.

Graph and vector retrieval

Multi-Hop GraphRAG

Follow evidence across documents.

I combined graph traversal and vector search to connect evidence across documents. Compare retrieval paths and inspect the passages behind each answer in the interactive demo.

Pipeline map

SourceSynthetic documents
ConnectEntities + relationships
RetrieveKuzu + LanceDB
InspectPaths + source passages

View repository Read the approach

Scope & evidence

Source: An independent project with 11 synthetic documents. The demo benchmark contains three questions.
Data flow: Document passages feed a Kuzu relationship graph and a LanceDB vector index. The app exposes retrieval paths and supporting passages for inspection.
Default: The default demo uses hash embeddings and heuristic answer extraction. Its answers are not generated by a language model.
Limits: The small synthetic benchmark illustrates the retrieval approach. It does not establish general accuracy or production performance.

Retrieval authorization

Permission-aware RAG

See how access rules change retrieval.

I built a retrieval demo that filters records before ranking them. Change the tenant, role, and clearance settings to see which synthetic records become eligible for retrieval.

Pipeline map

SourceSynthetic hospital records
FilterTenant + role + clearance
RankLexical cosine similarity
InspectAllowed source passages

View repository Read the approach

Scope & evidence

Source: An independent demonstration using synthetic hospital records, with no patient data.
Data flow: Python filters records by tenant, role, and clearance before ranking the eligible passages with lexical cosine similarity.
Backend: The repository also contains a separate FastAPI backend using PostgreSQL row-level security, pgvector, and tenant audit chains. That backend is not running in this demo.
Limits: Streamlit settings simulate access rules. The demo does not authenticate users or enforce database row-level security, and it is not a production security assessment.

More projects

Other builds

Healthcare data engineering

NY SPARCS

I used Python and DuckDB to normalize New York’s annual discharge files and build Parquet reporting marts. The app explores sample data, not statewide totals.

View repository

Operational data modeling

ED Operations

I modeled Scotland’s emergency-department data by site and month in Python and DuckDB, then compared their results. The forecast’s uncertainty interval includes zero improvement.

View repository

Customer analytics

Ecommerce Retention & Growth

I built subscription features using data available before each prediction cutoff. Includes a real-data sample and a separate synthetic demo.

View repository

Financial modeling

AutoModeler

Financial data to Excel models with linked formulas. Stops when a balance-sheet adjustment exceeds the configured limit.

View repository

Text classification

Ticket Intel

TF-IDF and Naive Bayes for ticket classification. The app uses synthetic examples; code and evaluation are on GitHub.

View repository

Model evaluation

SignalForge

I compared churn models on IBM Telco data and documented how I evaluated them.

Read the case study

My approach

How I work

I build pipelines I can explain: where the data came from, how it changed, and how I checked the result.

Source
Understand.
I start with where it comes from, what each row means, what’s missing, and how often it changes.
SPARCS: annual schema changes
Model
Build.
I define the output before building the pipeline. Then I keep the extraction, staging, and transformation steps easy to follow.
ED Operations: the site-month model
Checks
Check.
I compare keys, counts, and totals against the source. If something doesn’t line up, the check should fail where I can see it.
ED Operations: SQL reconciliation tests
Hand-off
Document.
I document the sources, field definitions, refresh steps, and limits so someone else can pick it up without guessing.
SPARCS: source, build, and demo instructions

Background

Experience

My background is in healthcare reporting and business intelligence. That’s where I’ve worked on turning operational data into reports people can use.

Independent practice

CGC Labs

Founder / Data engineering

I started CGC Labs for my independent data engineering work. I focus on getting data out of legacy systems, cleaning it up, checking it, and documenting how it works.

Professional experience / McCook, Nebraska

Community Hospital

Business Intelligence Analyst

I built Tableau dashboards and automated recurring reports. I also created SQL staging for an electronic medical record migration and validation checks for billing data.

I keep this work high-level here. Employer code, internal systems, and patient data stay private.

Beyond this portfolio

Elsewhere.

Independent practice

CGC Labs

My independent data engineering work and consulting.

cgclabs.org

Product in progress

MeasureMap

I’m building a registry for metric definitions, owners, formulas, and source references.

measuremap.cgclabs.org

Contact

Let’s talk.

Hiring a data engineer or need help with a project?
Tell me what you’re working on.

Email mechristian.g.callahan@gmail.com Find me on GitHub

Dataengineer.

Projects

Other builds

NY SPARCS

ED Operations

Ecommerce Retention & Growth

AutoModeler

Ticket Intel

SignalForge

Understand.

Build.

Check.

Document.

Education

CGC Labs

MeasureMap

Let’s talk.

Dataengineer.

Projects

Other builds

NY SPARCS

ED Operations

Ecommerce Retention & Growth

AutoModeler

Ticket Intel

SignalForge

Understand.

Build.

Check.

Document.

Education

CGC Labs

MeasureMap

Let’s talk.

Data
engineer.

Data
engineer.