# Moe Chizari at the Census Bureau: data science training and the move from SAS to Python

U.S. Census Bureau, 2020 to 2024. Last changed 22 October 2026. Other roles: https://chizari.me/work/

From 2020 to 2024 I was Data Science Transformation Lead at the U.S. Census Bureau, working through the Orus Group contract. The engagement ran in the same years as my LMI role. Facts and figures about the Bureau and its programs come from the public documents listed at the end.

## What I did

I delivered and scaled the Bureau's Data Science Training Program across both of its cohort paths, the Data Science Generalist Path and the Machine Learning Specialist Path [1]. The program is part of the Bureau's Statistical Data Modernization project [1]. My record includes a U.S. Census Bureau Certificate of Appreciation for the 2022 Data Science Training Program, from the Chief Learning Officer and the Program Lead. Under Commerce policy, a certificate of appreciation is the form of recognition open to contractors [2] [3].

I architected the migration of statistical production workloads from SAS to Python. The work was aligned to the Bureau's published commitment, in October 2022, to "transition legacy systems to open-source where possible" and to "require open-source for establishing and developing new IT systems" [4]. I established standards for statistical analysis, time-series modeling, automated reporting and MLOps. I wrote about moving from SAS to Python in October 2024: [Sas2py: moving from SAS to Python](https://chizari.me/projects/sas2py/).

I led research and development on generative AI and LLM evaluation, and built interactive Plotly Dash dashboards for senior leadership. I mentored junior and mid-level data scientists and established enterprise code-review practices.

## What the public record shows

The training program began before my engagement, as a pilot. Its first cohort started in January 2020 with 51 of 125 applicants accepted, on a data science generalist path and a machine learning path, and graduated in June 2020 [5]. The Bureau now describes a five-month cohort for current full-time employees, with online coursework, live instruction, capstone projects and mentorship, alongside a self-paced On-Demand option [1].

The program fits the Federal Data Strategy. Action 4 of its 2020 Action Plan asked CFO Act agencies to identify the data skills they need, assess current staff, run a gap analysis and act to close the gaps, with milestones running to 31 December 2020 [6]. The CDO Council's Data Skills and Workforce Development Working Group published the Census pilot as a case study for agencies carrying out that action in FY2021 [5].

The move away from SAS fits the Bureau's own modernization plan. In October 2022 the Senior Advisor to the Deputy Director for IT and Operations wrote that the Bureau's IT portfolio rested largely on closed-source, proprietary technologies. The same post named the cloud-based Enterprise Data Lake "the central hub" of modernization and said the Bureau would need new cloud and data science skills [4]. In January 2023 the Director wrote that AI and machine learning were "facilitating fuller development and automation" of economic statistics built from administrative data [7].

In April 2024 GAO reported that the Bureau had begun four modernization programs for systems that collect, process and disseminate data. The program office's life-cycle cost estimate for the Enterprise Data Lake was $337 million, covering 2021 to 2033 [8]. GAO found the cost and schedule estimates for the dissemination program, CEDSCI, unreliable and made five recommendations, and Commerce agreed with them [8]. These documents describe the Bureau's programs, not my work.

## The workforce in OPM's public data

OPM's Federal Workforce Data counts the Census Bureau's employees at each fiscal year end. The shaded band marks my years, 2020 to 2024. These are public counts for the whole agency, not figures from my work. [11]

Chart, headcount: Census Bureau employees in OPM's September snapshots, 2014 to 2025, and in August 2026, the newest file: 14,148 in September 2014, 21,890 at the high in September 2020, 10,935 in August 2026. [11]

Chart, hires and departures: Hires (accessions, transfers in included) and departures (separations of every kind) by fiscal year, counted by the date each action took effect. In fiscal year 2025, 2,610 joined and 4,310 left, 258 of them under the Deferred Resignation Program. [12]

| Snapshot | Employees |
|---|---|
| September 2014 | 14,148 |
| September 2015 | 15,814 |
| September 2016 | 14,249 |
| September 2017 | 16,280 |
| September 2018 | 14,814 |
| September 2019 | 20,947 |
| September 2020 | 21,890 |
| September 2021 | 18,170 |
| September 2022 | 13,744 |
| September 2023 | 15,348 |
| September 2024 | 13,232 |
| September 2025 | 11,288 |
| August 2026 | 10,935 |

| Fiscal year | Hires | Departures |
|---|---|---|
| FY2015 | 5,655 | 5,143 |
| FY2016 | 3,439 | 3,312 |
| FY2017 | 5,211 | 3,789 |
| FY2018 | 2,808 | 3,677 |
| FY2019 | 8,205 | 3,285 |
| FY2020 | 7,982 | 7,673 |
| FY2021 | 6,525 | 10,275 |
| FY2022 | 3,743 | 5,927 |
| FY2023 | 6,458 | 5,415 |
| FY2024 | 2,804 | 3,822 |
| FY2025 | 2,610 | 4,310 |

## Timeline

- **January 2020:** The pilot's first cohort begins, with 51 of 125 applicants accepted. [5]
- **2020:** My engagement through the Orus Group contract begins.
- **2022:** Certificate of Appreciation for the 2022 Data Science Training Program.
- **5 October 2022:** The Bureau commits to moving legacy systems to open source where possible and to requiring open source for new IT systems. [4]
- **29 April 2024:** GAO-24-105979 on the Bureau's four modernization programs. [8]
- **2024:** My engagement ends.
- **2 September 2026:** The Bureau's training page, last revised that day, lists both cohort paths. [1]

## Sources

1. [U.S. Census Bureau, Lifelong Learning Opportunities: Data Science Training Program](https://www.census.gov/about/census-careers/working-at-census/learning-opportunities.html)
2. [U.S. Department of Commerce, Performance Management Handbook (Recognition), Chapter 14](https://www.commerce.gov/hr/resources/handbooks/performance-management/recognition/chapter-14)
3. [U.S. Department of Commerce, Contractors and awards (2017 to 2021 archive)](https://2017-2021.commerce.gov/node/1088.html)
4. [U.S. Census Bureau, Technology Transformation at the Census Bureau: Building a Modern, Data-Centric Ecosystem (5 October 2022)](https://www.census.gov/newsroom/blogs/research-matters/2022/10/technology-transformation.html)
5. [CDO Council, Data Skills Training Program Case Studies: Census Bureau Data Science Training Pilot](https://resources.data.gov/assets/documents/CDOC%20Data%20Skills%20Case%20Studies%20v6.pdf)
6. [Federal Data Strategy 2020 Action Plan](https://strategy.data.gov/assets/docs/2020-federal-data-strategy-action-plan.pdf)
7. [U.S. Census Bureau, A Look Ahead: 2023 (Director's blog, 23 January 2023)](https://www.census.gov/newsroom/blogs/director/2023/01/a-look-ahead-2023.html)
8. [GAO-24-105979, IT Modernization: Census Bureau Needs Reliable Cost and Schedule Estimates (29 April 2024)](https://www.gao.gov/assets/gao-24-105979.pdf)
9. [U.S. Census Bureau, Statistical Abstract of the United States: 1977 (98th edition)](https://www.census.gov/library/publications/1977/compendia/statab/98ed.html)
10. [R documentation, US State Facts and Figures (datasets package: state.x77, state.region, state.abb), naming the Census Bureau's 1977 Statistical Abstract and County and City Data Book as the source](https://stat.ethz.ch/R-manual/R-devel/library/datasets/html/state.html)
11. [OPM, Federal Workforce Data: employment snapshots, September 2014 to September 2025 and August 2026](https://data.opm.gov/) (read 22 October 2026)
12. [OPM, Federal Workforce Data: accessions and separations files, October 2014 to August 2026](https://data.opm.gov/) (read 22 October 2026)

Interactive demo on this page: Try it: SAS to Python on Census Bureau data. A teaching converter turns a SAS program into pandas code, then computes the SAS listing and the pandas output in your browser from Census Bureau figures for the 50 states, published in 1977, so you can check that the two agree. [9] [10]

Agency names are where the work was done. They are not endorsements.