building AI that admits uncertainty
I build AI systems that know when they're wrong.
Founder of Fact AI Lab · AI Researcher & Applied ML Scientist · Content Creator
From spreadsheet firefighting to audit-ready AI - helping small and mid-cap companies deploy eval-first systems that pass audit and hold up in production.
- •CFOs and finance leaders automate month-end close with reliable AI
- •Heads of Data & Analytics build audit-ready ML pipelines
- •Product teams in regulated industries ship AI features that pass compliance
20+ peer-reviewed publicationsh-index 9PhD Computer ScienceAI Command Room newsletterHow I Can Help
Automate Workflows with Reliable AI
Stop firefighting manual processes, AI slop, or hallucinations. As a senior advisor, I help teams scope and ship reliable AI in high-stakes workflows - starting with the ones that cost the most time.
Learn to Build AI That Works
Writing, videos, and frameworks on LLM reliability, RAG systems, and deploying AI in regulated environments.
Apply Research to Real Problems
20+ peer-reviewed publications on uncertainty estimation, autonomous systems, and LLM reliability. Open-source code for most projects.
Engagement Options
AI Workflow Accelerator
6–8 week engagement
Identify high-ROI workflows → build → validate → deploy with eval framework and audit trail.
RAG Audit Readiness Assessment
3–4 week assessment
Retrieval quality, calibration gaps, hallucination risk, and compliance exposure - scored and prioritized.
AI Strategy Sprint
2-week structured sprint
Map AI opportunities → prioritize by ROI + risk → leave with a scoped first project and governance framework.
Who This Is For
Revenue:$10M–$500M companies
Roles:CFO, VP Finance, Head of Data, FP&A Lead, VP Engineering
Industries:Financial services, insurance, legal, healthcare - any regulated vertical
Stage:You have data and some AI exploration, but need reliable production deployment
The RAMP Framework
How I approach every AI deployment:
RReview
Audit existing workflows for AI-readiness and compliance requirements.
AAssess
Measure data quality, identify failure modes, set evaluation baselines.
MMap
Match the right AI approach (LLM, RAG, agents, classical ML) to each workflow.
PProve
Deploy evaluation-first - measure before you ship, monitor after you launch.
If You're New Here
Why Your RAG Pipeline Fails at Retrieval, Not Generation
Most RAG debugging focuses on the LLM. The real problem is almost always in the retrieval step - here's how to diagnose and fix it using productionvalidated tec
2026-03-10 · 7 min read
Python Type Hints That Actually Catch Bugs in Production
Most Python type hint guides show you the syntax. This one shows you the patterns that catch real bugs before they reach users.
2026-02-20 · 4 min read
Free Resources
ML Quick Review (PDF)Free
Core ML concepts from linear models through deep learning. One-page reference for interviews and project kick-offs.
Multi-Provider Proxy Agent (PDF)Free
Architecture and implementation guide for building a multi-provider LLM proxy - routing, fallback, and load balancing.
SCQA Slide TemplateFree
Consulting-style Situation–Complication–Question–Answer framework for high-stakes presentations. Copy to Drive and customize.
Latest Videos
Latest Writing
Latest Research
Recommender System for Data Science Learning and ResearchInternational Journal of Artificial Intelligence in Teaching and Learning (IJAITL) · 2025