About Precision Federal
Precision Federal is a small business building AI, data and software systems for government agencies. The company’s founder loves God, and the company is built on Christian faith: every number we publish is real, every claim we make is verifiable, and the work is done with excellence as service to God. Faith is welcome here and openly shared.
How this role works
- You join the bench. You apply with your weekly hours and start time. We set your hourly rate from your résumé, and we may name you on proposals that match this role. If we don't win, you owe nothing.
- A contract is awarded. The award funds the team named in the proposal. Your engagement begins at the hourly rate we agreed on.
- You work remote, paid from the contract. ~120–600 hrs per funded award — and if no award ever funds, you owe nothing and lose nothing.
What you'll do
- Design how a graph-structured model is partitioned, trained and run across nodes with intermittent or partial connectivity
- Work out what is provably recoverable when data at each node is sparse, and say plainly where the method stops holding
- Build the measurement harness and honest error bars, aleatoric and epistemic, that let a government research team rerun the results in house
- Tell us where a learned model should be combined with expert-stated rules, and how to check that the combination actually helps
Required qualifications
- Real depth in graph neural networks, distributed or federated learning, neuro-symbolic methods, or uncertainty quantification, shown through peer-reviewed publications, patents, or a shipped system with numbers behind it
- Hands-on experience implementing and evaluating models in PyTorch, JAX or a comparable framework, including graph libraries such as PyTorch Geometric or DGL
- U.S. citizen or lawful permanent resident (U.S. person)
- Willing to be named on a federal proposal with a one-page resume, and to sign a short letter of commitment
Preferred qualifications
- PhD or equivalent research record in machine learning, statistics, applied mathematics or a related field
- Decentralized inference, message passing, or learning under bandwidth and connectivity limits
- Prior work on federally funded research, including writing or serving as key personnel on a proposal
- Experience combining learned models with symbolic, rule-based or physics-based knowledge
Compensation & structure
1099 consulting engagement. $70–$145 per hour; we set your rate from your résumé and experience. Hours are funded by federal contract awards — this seat is scoped at roughly 120–600 hours per funded award, fully remote. Work and pay begin only if the contract is awarded — no award, no hours, no obligation.