I build the LLM and data platforms behind the product, and I’ve led the teams that own them.

Director of Data Engineering & AI

Tastewise

Tech Lead → Team Lead, Application → Data & AI Team Lead → Director of Data Engineering & AI

Started on the core ETL platform, built out the company’s first proper data platform, and grew into owning the data products and the whole AI/agent layer.

• Built and have owned the LLM pipeline behind the product’s AI features since 2023 — pull data from Elasticsearch, ground a versioned prompt in it, call the model, and turn the result into imagery with genai — with runtime JSON-schema validation that feeds errors back to the model for automatic repair, back when getting valid JSON out of an LLM was still hard.
• Cut the wait from a ~90-second blank screen to answers that appear as they generate: moved the LLM path onto WebSocket token streaming with optimistic partial-JSON parsing, so the UI renders partial results instead of holding for the full response.
• Added an LLM router that classifies each request and routes it to the right flow, chained dependent calls, and built the operational scaffolding around them — PromptLayer for prompt versioning (which I evaluated and introduced; still in use) and Langfuse for tracing the chains.
• Stood up internal AI-platform infrastructure: MCP servers exposing the company’s data platform to agents, an OAuth2 AI gateway that mints per-user tokens, and a Slack knowledge-sharing agent owning the team’s LLM wiki.
• Led an internal platform that lets anyone at the company build sub-apps inside the product — no R&D deployment or security burden. With my AI engineers I built the MCP layer giving the Lovable coding agent governed access to Tastewise: a system prompt, progressive-disclosure API tools (BM25 over our OpenAPI), a Storybook MCP for the design system, and the PromptLayer MCP — with skills delivered as markdown tool-outputs where Lovable can’t take them over MCP — all behind OAuth2 through our AI gateway.
• Introduced Claude Code to R&D in July 2025 and drove its adoption — within a couple of months the whole org was building with it — then led the build of an internal Claude Code plugin packaging the company’s shared skills and MCP servers.
• Moved the data platform onto Databricks and Delta Lake over 292 commits, rewriting jobs that used to run for days into PySpark on Airflow, so they finish in hours instead.
• Built the ingestion behind 1M+ indexed restaurants & venues, unifying 10 mismatched collectors into one schema and orchestrating the jobs on Airflow with KubernetesPodOperators (EKS).
• Kicked off the Angular 12 → React 18 migration, writing an interop layer that rendered React components inside the live Angular app for an incremental cutover — widening the hiring pool and lifting delivery velocity.
• Started Tastewise’s design system with the designers in 2023 and built its first Storybook — later the basis for the Storybook MCP that lets the internal app-builder’s agent build with the real components.
• Built company-wide SSO by hand with Passport.js — no managed auth provider — standing up authentication across the product’s services.
• Led the observability overhaul, moving the platform from Logz.io to Datadog and standardizing a custom Node.js & Python logger across every service, so following a request went from grepping scattered logs to a single query — cutting bug investigations from hours to minutes.
• Modernized the backend across two migrations: TasteGPT from Flask to FastAPI, and the Nodejs scrapers from TypeScript to a fully async Python framework.
• Grew the team from 2 to 9 engineers across full-stack, data, AI and data science, built the interview process we still use, and took over the prior team’s ML pipelines with zero handover.

• Python
• Databricks
• PySpark
• Airflow
• Kinesis
• Firehose
• Lambda
• Pydantic
• MCP
• Langfuse
• FastAPI
• Kubernetes (EKS)
• Datadog
• TypeScript

Full-Stack Team Leader

Zencity

Architected the civic-engagement (B2G) product’s survey-submission platform, and grew from developer to leading the full-stack team.

• Cut survey-collection cost to near zero by replacing a third-party SaaS with an in-house platform — a React client plus a pipeline of event-driven Lambda microservices over SQS, fanning each submission through ingestion, score aggregation, ML entity-enrichment (BERT) and location enrichment before persistence — running across ~100 cities at thousands of submissions a day.
• Championed and drove the Node.js codebase migration from JavaScript to TypeScript, sharply reducing production bugs, and added a schema layer over MongoDB that further hardened system stability.
• Operated, maintained and monitored the Confluent Kafka CDC (Change Data Capture) pipeline streaming MongoDB’s oplog into Elasticsearch — thousands of events a day fanned across 50 consumer groups, partitioned by a modulo on the Mongo ObjectId.
• Built Zencity’s first microfrontend — a host-and-remotes Webpack Module Federation architecture — upgrading Webpack and wiring the remotes to load and communicate with each other both locally and across staging and production.
• Built device fingerprinting and throughput optimization into the submission flow, hardening it and making it faster to onboard new cities.
• Grew and led the full-stack team from 2 to 7 engineers after being promoted from developer.

• TypeScript
• React
• Node.js
• AWS Lambda
• SQS
• Kafka (Confluent)
• MongoDB
• Elasticsearch
• Webpack Module Federation

Co-Founder & VP R&D

Snipe

Co-founded an esports startup and ran R&D, shipping ML-driven products for League of Legends players.

• Built Sightstone, which predicted favorable League of Legends matchups by pre-ingesting millions of games — Spark/Scala for processing, scikit-learn for the models, MongoDB for storage.
• Built Matchmaker, a mobile app that profiled players’ playstyle to recommend teammates and streamers, tackling game loneliness and content discovery.
• Grew to 100k+ downloads and 10k+ weekly active users across 4 products in two years, raised a pre-seed round, and moved the company to Silicon Valley.

• React
• React Native
• AWS
• MongoDB
• Spark (Scala)
• scikit-learn

Data Developer Team Leader

Israeli Military Intelligence — Unit 8200

Led a data team building tools that surfaced new insight from previously-inaccessible data.

Israel’s elite SIGINT unit, the talent pipeline behind Check Point, Palo Alto Networks and Wiz.

• Built tooling that made previously-inaccessible mission-critical data queryable, unlocking insights from data the organization already held.
• Led a team of 5 data developers and analysts across seniority levels.