We source, vet, and manage hiring so you can meet qualified candidates in days, not months. Strong English, U.S. time zone overlap, and compliant hiring built in.
.svg){kind=logo}
Turing.jl is the probabilistic programming language for Bayesian machine learning in Julia. If you're building generative models, Bayesian inference systems, or causal inference pipelines, Turing is your framework. South connects you with senior probabilistic programmers from Brazil and Argentina who've shipped production Turing models for finance, pharma, and e-commerce. Get matched in days. Start your search with South today.
Turing.jl is an open-source probabilistic programming language built on Julia. It's designed for Bayesian inference and generative modeling. With Turing, you define a probabilistic model using a simple syntax, and Turing automatically handles inference via Markov Chain Monte Carlo (MCMC), variational inference, or other algorithms.
Turing is used for generative models, causal inference, Bayesian regression, and mixture models. The language is particularly strong in quantitative finance (risk modeling, portfolio optimization), pharmaceutical research (clinical trial analysis), and e-commerce (demand forecasting, recommendation systems). Turing abstracts away the complexity of MCMC and variational inference, letting you focus on modeling.
GitHub shows 2,000+ stars and active development by the MIT probabilistic computing lab. Julia's adoption in LatAm is growing, especially in Brazil where fintech companies (Nubank, Inter) and academic labs use Turing for risk and causal inference. Turing talent is rarer than Python ML talent, but LatAm has a strong emerging community.
Turing is not a general-purpose language. It's specifically for probabilistic modeling. If you're doing traditional supervised learning or deep learning, Turing is overkill. But if you need to express uncertainty, reason about causal relationships, or sample from distributions, Turing is the right tool.
Hire a Turing expert when you're building Bayesian inference systems or generative models. Common scenarios: you need to forecast demand with uncertainty quantification. You're running A/B tests and need Bayesian sequential analysis. You're building causal inference models for healthcare or policy. You need to quantify risk in financial portfolios.
Don't hire for Turing if you're doing standard supervised learning. If your problem is classification or regression with point estimates, a linear model or XGBoost is simpler and faster. Turing is for problems where you care about uncertainty, heterogeneous effects, or causal structure.
Ideal team structure: one senior Turing engineer (5+ years probabilistic programming), one data scientist who works on modeling, and one DevOps engineer managing compute infrastructure for MCMC chains. For smaller teams, a senior Turing engineer can handle both modeling and system work.
Turing shines in quantitative industries: fintech, pharma, insurance, academic research. If you're building a consumer application, you probably don't need Turing.
A strong Turing engineer understands Bayesian inference deeply. They know the difference between MCMC algorithms (Metropolis-Hastings, Hamiltonian MC), when to use variational inference vs MCMC, and how to diagnose convergence issues. They've shipped Turing models to production and debugged inference problems.
Red flags: engineers who don't understand Bayesian statistics. Turing is a tool for Bayesian modeling; if they can't explain posterior distributions or prior selection, move on. Also watch for overconfidence in inference results. Good Turing engineers run diagnostics (Rhat, effective sample size) and validate models rigorously.
Junior (1-2 years): Understands Bayesian fundamentals. Can write simple Turing models with guidance. Knows one inference algorithm well (e.g., Hamiltonian MC). Needs mentorship on model specification and diagnostics.
Mid-level (3-5 years): Comfortable designing complex Bayesian models. Knows trade-offs between inference algorithms. Can troubleshoot convergence issues and diagnose problems. Understands hierarchical modeling. Can integrate Turing with data pipelines.
Senior (5+ years): Has shipped production Turing systems. Understands Bayesian inference at a theoretical level. Can design efficient models that converge quickly. Knows when Turing is the right choice and when simpler methods suffice. Has experience with large-scale inference and distributed sampling.
For remote work, Turing engineers in LatAm are typically UTC-3 to UTC-5, giving you 5-8 hours of overlap with US teams. Soft skills: they should be able to explain Bayesian reasoning to non-technical stakeholders.
Tell me about a Bayesian model you built to solve a business problem. Listen for: problem framing, prior selection, inference method, and model validation. Strong answers mention checking prior predictive distributions and posterior diagnostics.
You've built a Turing model that isn't converging. Walk me through your debugging process. Good answers start with checking Rhat values, effective sample size, and trace plots. They might try reparameterization, better priors, or different samplers. They should mention specific Turing diagnostics.
Describe a time you explained Bayesian results to a non-technical stakeholder. Listen for: clarity, avoiding jargon, communicating uncertainty appropriately. A strong answer mentions translating posterior distributions into business metrics.
When have you chosen not to use Bayesian methods and used something simpler instead? Maturity signal. A great answer: 'We started with Turing, but realized point estimates were sufficient and switched to logistic regression for speed.' Shows understanding of trade-offs.
How do you stay current with Bayesian modeling and Turing development? Strong engineers follow Turing releases, read Bayesian modeling papers, and engage with the Julia community.
Explain the difference between Markov Chain Monte Carlo and Variational Inference. When would you use each? Evaluation: they should understand that MCMC is slower but asymptotically correct, VI is faster but approximate. They should know when each is appropriate (MCMC for final inference, VI for fast prototyping).
You're building a hierarchical model in Turing with student-level data nested in schools. How would you structure the model? Evaluation: they should understand partial pooling, random intercepts/slopes, and how Turing handles hierarchical specifications. A strong answer mentions prior specification at each level.
What's the Rhat diagnostic and why does it matter? Evaluation: they should know that Rhat measures MCMC convergence (values under 1.01 are good), and they should understand what to do if Rhat is high (more iterations, reparameterization, better priors).
How would you implement a mixture model in Turing? What challenges might you encounter? Evaluation: they should understand label switching, how to specify discrete mixture components, and computational challenges. Strong answers mention strategies to handle label switching.
Explain prior specification in Bayesian modeling. How do you choose priors? Evaluation: they should understand that priors encode domain knowledge, weak priors are conservative, strong priors can guide inference. They should mention prior sensitivity analysis.
You have customer spending data with multiple observations per customer. Build a Turing model that estimates individual-level effects with partial pooling. Include prior specification and convergence diagnostics.
Scoring: model structure (40%), prior selection (20%), inference and diagnostics (30%), clarity (10%). Strong submissions show understanding of hierarchical modeling and Bayesian diagnostics.
LatAm Rates (2026):
US Market Comparison:
Turing engineers are rare, so salaries are premium relative to other languages. LatAm Turing talent is concentrated in Brazil (São Paulo, Rio), where quantitative finance and research labs use Julia and Turing. Argentina has emerging Turing adoption in fintech and academic institutions.
Brazil has a strong quantitative finance community where Turing adoption is growing rapidly. Engineers at Nubank, Inter, and other fintechs use Turing for risk modeling and portfolio optimization. The Julia community in LatAm is small but tight-knit and high-quality. Research institutions (USP, UNICAMP, UNAM) are increasingly using Turing for academic work.
Time zone alignment is valuable for Bayesian modeling work. Most LatAm Turing engineers are UTC-3 to UTC-5, giving you 5-8 hours of real-time collaboration with US teams. For complex modeling problems, synchronous collaboration with your Turing engineer is invaluable.
English proficiency is high among LatAm Bayesian modelers. They've learned Turing, Julia, and statistical theory through English-language documentation and research. Communication about inference challenges is clear and precise. Cultural alignment: LatAm Bayesian modelers are typically research-focused and intellectually rigorous.
Cost efficiency is exceptional. You're saving 45-50% on a LatAm Turing engineer compared to US rates. For a niche skill like probabilistic programming, this ROI is outstanding.
Tell us about your Bayesian modeling problem: are you building generative models, causal inference systems, or Bayesian analytics? We match from our pre-vetted network of LatAm probabilistic programmers, filtering for Bayesian inference expertise, Turing experience, and seniority level. You interview 2-3 candidates in 48-72 hours. We handle ongoing support: if the engineer isn't working out, we replace them within 7 days at no additional cost. Our 30-day guarantee ensures the right fit or your money back.
South's vetting includes Bayesian inference assessments, probabilistic programming exercises, and model diagnostics knowledge. We verify production experience and ask about inference challenges they've solved. This filters out candidates who've only done academic Turing projects.
Once matched, you get a fully integrated engineer with visa sponsorship, equipment, and compliance handled. Start matching with Turing experts today.
Turing is used for Bayesian inference, generative modeling, causal inference, and uncertainty quantification. Any problem where you need to reason about probability distributions is a candidate for Turing.
All three are excellent probabilistic programming frameworks. Turing is strongest in Julia ecosystems and high-performance computing. PyMC is more mature and has more examples. Stan is industry-standard for certain domains. Choose based on your infrastructure and team expertise.
Turing is for interpretable Bayesian generative models where you understand the data generating process. Deep learning is for high-dimensional data where you don't have strong priors. They solve different problems. Use both when appropriate.
Senior Turing engineers in LatAm cost $105,000-$140,000/year, roughly 45-50% less than US rates for equivalent talent. Brazil is the primary source.
You'll interview qualified candidates within 48-72 hours of describing your needs. Most placements finalize within 1-2 weeks. Turing talent is rare, so we prioritize speed.
For custom Bayesian model development, hire mid-level or senior (3+ years probabilistic programming). For using existing Bayesian models, a junior engineer can handle most work.
Yes. South places engineers for both full-time roles and project-based engagements (3-6 months). Rates adjust based on engagement type.
Most LatAm Turing engineers are in Brazil (UTC-3), giving 5-8 hours of overlap with US teams. This is ideal for collaborative Bayesian modeling work.
We assess Bayesian inference knowledge, ask about production model diagnostics, and request probabilistic programming samples. We verify their understanding of MCMC and variational inference.
We offer a 30-day guarantee. If the engineer doesn't meet expectations, we replace them at no additional cost. We solve fit issues via intensive onboarding and clear modeling objectives.
Yes. South handles visa sponsorship, payroll, tax compliance, benefits, and equipment. One all-in monthly fee; we manage everything.
Absolutely. We've placed small teams of 2-3 probabilistic programmers on inference projects. We ensure team alignment on Bayesian methodology and modeling standards.
