ABOUT

AI LEAD, JFROG

Liran Baba

Somewhere between "AI will change everything" and "ok but who's going to actually build and deploy this."

Pixel portrait of Liran Baba

18 years in enterprise tech. I lead AI on JFrog's CIO team. Day to day that looks like evaluating tools, building internal platforms, and working with engineering and security to get AI into actual workflows. Less "art of the possible," more "will this pass a security review."

I got here through a long stretch at Avaya, where I architected solutions for Dell, Apple, Boeing, and Comcast, and built tools for speech analytics and knowledge management. That background is probably why I care more about integration headaches than hype cycles.

I also mentor people on AI adoption and build things on the side, mostly productivity tools and automation.

PROJECTS

001

GolemPII-v1

TYPE
ML Model
STATUS
Released

Hebrew PII detection and redaction model.

No dedicated Hebrew PII detection model existed for local deployment. GolemPII fills that gap with a fine-tuned model trained on 115K Hebrew examples, enabling privacy compliance with Israeli data protection regulations without sending data to external APIs.

MLNLPTransformersHebrew
002

Claudoscope

TYPE
macOS App
STATUS
Active

Cost visibility and session analytics for Claude Code.

Claude Code users had no way to track spending or analyze usage patterns across sessions. Claudoscope gives them a native macOS dashboard for real-time cost monitoring.

macOSSwiftClaude Code
003

Adapt-CV

TYPE
Web App
STATUS
Active

Align resumes with job descriptions for better ATS pass rates.

Job seekers struggle to tailor resumes for each application. Adapt-CV analyzes the gap between a resume and a job description, then suggests targeted improvements to increase ATS compatibility.

StreamlitClaude APIPython
004

Datasets


004a

CustomerPersonas

250K synthetic customer personas across six industries, ready for training and evaluating customer service AI systems.

Dataset Python NLP
004b

GolemGuard

Comprehensive Hebrew language dataset for training and evaluating PII detection and masking models. Contains 115K synthetic text samples representing various document types and communication formats commonly found in Israeli professional and administrative contexts.

Dataset Hebrew NLP
005

Reddit MCP

TYPE
MCP Server
STATUS
Active

MCP server for searching and reading Reddit.

AI agents have no way to access Reddit. This server gives MCP-compatible agents 13 tools for browsing posts, threads, user profiles, and subreddit wikis. Credentials rotate automatically with per-key rate limiting.

PythonMCPAsyncPRAW
006

horita

TYPE
macOS App
STATUS
Active

Your next meeting and a countdown in the Mac menu bar.

Seeing what's next meant opening a calendar and digging for the call link. horita shows the next meeting with a countdown in the menu bar and joins Zoom, Google Meet or Teams in one click. It reads the calendars already on your Mac, and the data never leaves it.

macOSSwiftEventKit

WRITING

  1. Everyone can ship to production now. Nobody is on call.

    75% of CISOs have found unsanctioned AI tools already running in production. Non-developers are shipping real software, and no one has staffed day 2.

  2. I gave Claude Code read-only access to its own spending

    Claudoscope 1.0 ships a read-only MCP server so Claude Code can query your own session costs, four cost-alert rules, and two fixed transcript parsing bugs.

  3. The hard part of a self-service agent platform isn't the agents

    Gartner expects 40% of enterprise apps to ship AI agents by end of 2026. After building AgentHive at JFrog, I think generating the agent is the easy part.

All posts →

CONTACT

What are you building?

I like connecting with people who ship things. Say hi.