ahiya.dev
Ahiya Butman
Clearest at the seam between the work and the people it’s for.
In plain terms: I build production AI systems for research, analysis, and automation, and I ship them end to end, from intake to delivery.
AI Engineer · Founder · Independent Builder
I founded StatViz, an AI-augmented analysis platform serving academic institutions in Israel, Herzog College among them.
Before StatViz, software engineer in Unit 8200, and a BSc in Computer Science, AI focus, from the Open University of Israel.
Day to day: Python · TypeScript · Next.js / React · FastAPI · PostgreSQL · Docker · Claude and OpenAI APIs.
Selected work
StatViz
↗Founder & Lead Engineer · 2025–present
- Problem
- Researchers need rigorous statistics with defensible write-ups. An LLM on its own can't be trusted to compute, or even transcribe, a number.
- Built
- A seven-agent Claude pipeline (ingest → explore → analyze → report → deliver) around a sealed statistics engine: the model plans the analysis but never computes anything. Every number comes from a registry of validated Python functions, is provenance-stamped, and is re-verified in the final report. An independent model critiques each analysis plan before it runs; delivery is fail-closed behind human approval gates. Output: Hebrew RTL Word, interactive HTML, and PDF reports.
- Result
- Real clients, real reports, paid work: graduate students to academic institutions, Herzog College among them. 30+ supported tests, from t-tests to SEM and multilevel models.
Next.js · TypeScript · PostgreSQL / Prisma · Claude agent pipeline · Python (SciPy · statsmodels · pingouin · semopy)
stat-viz.com
HIT · AI Persona Research
Consulting · 2025
A Hebrew NLP pipeline for an academic study of youth sport dropout in Israel, built at Holon Institute of Technology. Synthetic personas, each defined by demographic factors (age, ethnicity, region, sport, training schedule, dropout reason), generate first-person Hebrew testimonials at scale, feeding a quantitative analysis of why Israeli teenagers leave organized sport. Built end to end: prompt engineering, generation, and output formatting for the research team's downstream pipeline.
Python · Hebrew NLP · LLM APIs · persona design
2L
↗Personal project · ongoing
A two-level agent orchestration system that breaks a vision into iterations and runs each through a pipeline of specialized agents (explorer, planner, builder, integrator, validator, healer), with a real-time observability dashboard.
Claude Code agents · Bash · Python · JSONL event log · HTML dashboard
github.com/Ahiya1/2L
SelahOS
Personal · built, deployed, in daily use
A personal instrument panel for the ground layer of life: the daily rhythm of sleep, food, medication, hygiene, movement. No coaching, no streaks, no nudging. Concretely: the whole day is one mobile screen of tappable three-state circles; marking an anchor blooms the time it was met, then fades back to ground. Daily use takes under a minute.
Next.js 15 · React 19 · Supabase (Postgres, RLS, magic-link auth) · Tailwind · Vitest
What I can do for you
AI systems that ship
Agent pipelines, LLM workflows, and the evaluation and plumbing that keep them reliable once real users depend on them.
Internal tools & automation
Intake, processing, reporting, delivery: the systems that turn a manual service into a product. StatViz is this, running as a business.
Research engineering
Data pipelines, statistical analysis, Hebrew NLP. Work that has to be defensible, not just plausible.
Contact
If the work is unclear but important, write.
Open to full-time roles and contract work in applied AI, AI tooling, automation, backend, and research engineering. Email is fastest.
A desert story
The Sacred Potato: a novella, read as a walk.
Four parts · about an hour · best unhurried