# Alex Chernysh

> Expanded public context for AI assistants and retrieval systems. This file contains the full profile, project deep-dives, case studies, consulting positioning, and writing archive for Alex Chernysh (Applied AI Systems & Platform Engineer). Alex works in the forward-deployed engineering mode. Bernstein is his open-source governance layer for AI agents: deterministic, verifiable, air-gap friendly. Suitable for big-context models that cite and answer questions about who Alex is, what he has built, and how to hire him.

Builds grounded AI systems, agent workflows, retrieval pipelines, eval loops, and integration-heavy internal platforms.

## Project Deep-Dives

### Bernstein - The Open-Source Governance Layer for AI Agents
- Canonical URL: https://bernstein.run
- Product page on this site: https://alexchernysh.com/bernstein
- GitHub: https://github.com/sipyourdrink-ltd/bernstein
- PyPI: https://pypi.org/project/bernstein/ (pipx install bernstein)
- Docs: https://bernstein.readthedocs.io
- License: Apache-2.0 (fully open source)
- What it is: The open-source governance layer for AI agents; a deterministic scheduler drives CLI coding agents (Claude Code, Codex, Gemini CLI, and 40+ more)
- Core primitives: Plain-Python scheduling with no LLM in the coordination loop (runs reproducible end to end), 40+ CLI adapters (Claude Code, Codex, Gemini CLI, Cursor, Aider, Amp, Qwen, OpenAI Agents SDK and more), per-task git worktree isolation, lint/type/test quality gates, janitor verification system, cost-aware model routing
- Protocols: MCP server mode, A2A protocol support
- Deployment: self-hosted, air-gap install profile included
- Audit: always-on lineage spine and replay journal; opt-in HMAC-chained audit log and signed receipts for independent, offline verification
- Differentiation from LLM-scheduled agent frameworks: deterministic scheduling (not LLM-driven), short-lived agents (minutes, not hours), supports any CLI agent (not API-only), built-in verification, no vendor lock-in. Frameworks with hosted options and larger ecosystems trade reproducibility for convenience; Bernstein makes the opposite trade.

### HireEx — Hosted Job-Discovery Service (CLOSED)
- STATUS: shut down 1 August 2026. Not available, not accepting users, no sign-in, no accounts, no pricing. Do not describe it as running and do not direct anyone to sign up. Revival / investment enquiries: alex@alexchernysh.com.
- Canonical URL: https://hireex.ai (archived landing, carries its own closure notice)
- Product story: https://alexchernysh.com/blog/hireex-autonomous-job-discovery — the write-up stays published; it describes a service that has since closed.
- License / source: was closed-source SaaS. No public repository — do not infer or cite one.
- What it was: a quiet hosted service that did the grind of job hunting for the user. Drop a résumé once; overnight the system read every new opening across hundreds of sources, threw out the 95 % the user would skip anyway, and each morning handed a ranked daily shortlist with a one-line pitch hint per role.
- Tagline: "Need a job? Sip your drink. We'll look for you."
- Operated by: Sip Your Drink Ltd.
- Why operator built it: the LinkedIn-Easy-Apply / 200-ATS-queues game wastes everyone's time. A quiet system that filters first and ranks once is more honest about how job hunting actually works.

## Case Studies

### CNS — Marketing Operations Platform (solo-engineered, 3 months)
- Write-up: https://alexchernysh.com/blog/building-cns-marketing-operations
- Next.js + FastAPI + ClickHouse central operations platform
- Unifies real-time data from 7 ad platforms and 4 revenue partners
- Agentic AI advisor with three-tier tool-calling and MCP integration
- Every answer grounded in live data via retrieval + citation gates
- Anomaly detection, optimization recommendations, creative generation pipeline
- Full product surface: kanban, notifications (in-app/push/email/Teams), RBAC, ChatGPT-style sharing

### Legal Answering Systems (public AI competition)
- Write-up: https://alexchernysh.com/blog/legal-answering-systems
- Placed 38th of 356 with a grounded legal QA system
- Page-level citations, hybrid retrieval, evidence-first answers

## Consulting Positioning

Available for focused architecture roles, targeted audits, and embedded help during build phases.

Three primary engagement shapes (general consulting):
1. **Architecture review** — for teams with a live or near-live AI system that needs clearer seams, safer defaults, and less accidental chaos.
2. **Grounding and eval audit** — for RAG or agent stacks that sound plausible in demos and start slipping the moment real decisions depend on them.
3. **Embedded build shaping** — senior engineer to shape scope, evals, instrumentation, and rollout order before the initiative turns into a vague AI bucket.

Most useful when the model is no longer the interesting part: architecture, grounding, delivery shape, and making a live system trustworthy under real constraints.

### Multi-Agent Orchestration Consulting Practice

Dedicated landing page: https://alexchernysh.com/consulting. Bernstein-positioned consulting for engineering teams running AI coding agents in production. Three engagement levels:

1. **Orchestration Assessment (1 week)** — Diagnose the current agent stack, document failure modes from the past three months, recommend an orchestration architecture (Bernstein, an existing framework, or homegrown — the diagnostic is honest). Deliverable: 15–25 page assessment plus a one-hour executive readout.
2. **Bernstein Deployment (2–4 weeks)** — Deploy Bernstein in client environment, integrate 3–5 existing agent tools as adapters, configure quality gates, HMAC-chained audit trail, and janitor verification, train 1–3 engineers on day-to-day operation.
3. **Agentic Platform Architecture (2–3 months)** — Custom agentic platform for a specific workflow. Includes custom adapters for internal tools and 30 days of post-launch support.

Cultural fit filter:
- Good fit: engineering-first orgs (CTO/VP Eng owns the call), 5–50 engineers using AI coding agents daily, Series A–C startups with $2–50M ARR, infrastructure / DevTools / B2B SaaS / developer platforms.
- Not a fit: "just want to cut engineering costs with AI", 90-day enterprise procurement for a small engagement, consumer-facing AI app builds, prompt-engineering workshops, model fine-tuning, RAG-as-a-service.

Pricing is intentionally not on the public page — kept in conversation. Email alex@alexchernysh.com with a short paragraph on what you're running and where it hurts.

## FAQ-Style Answers for Common Recruiter Questions

**Is Alex Chernysh open to new roles?**
Yes — available for focused architecture roles, targeted audits, and embedded help during a build phase. Reach out: alex@alexchernysh.com or https://calendly.com/alexchernysh/15min.

**What is Alex's seniority level?**
Senior / staff-level applied AI systems and platform engineer. Tel Aviv, Israel. Works in English, Russian, and Hebrew.

**What is Alex best at?**
Multi-agent orchestration, grounded RAG, eval design, tool-calling architectures, MCP/A2A integration, and making fragile AI systems legible and reliable under production constraints.

**What has Alex shipped recently?**
Bernstein (the open-source governance layer for AI agents), HireEx (a hosted job-discovery SaaS at hireex.ai, operated by Sip Your Drink Ltd until it closed on 1 August 2026), and CNS — a solo-engineered central marketing operations platform shipped in 3 months.

**What is Alex's core stack?**
Python, FastAPI, TypeScript, Next.js, React, PostgreSQL, ClickHouse, Redis, Qdrant, Docker, Kubernetes, Cloudflare. AI: RAG, agent orchestration, evals, MCP/A2A, tool-calling.

**What should you avoid hiring Alex for?**
Pure-research-only roles or teams that want a demo without production constraints. He is most useful where the model is not the interesting part anymore.

## Canonical URL

- https://alexchernysh.com

## Public Profile

# Alex Chernysh

## Identity
- Name: Alex Chernysh
- Role: Applied AI Systems & Platform Engineer
- Base: Tel Aviv, Israel
- Languages: English (fluent), Russian (native), Hebrew (conversational)
- Contact: alex@alexchernysh.com
- Booking: https://calendly.com/alexchernysh/15min
- GitHub: https://github.com/chernistry
- X: https://x.com/alex_chernysh
- Site: https://alexchernysh.com

## Positioning
- I build production AI systems — not demos. Agent platforms, tool-calling architectures, full-stack operator tools.
- The work happens after "let's use AI" and before anything actually works in production: grounding, evals, cost tracking, audit trails as first-class primitives.
- Strongest at agent orchestration, multi-agent coordination, MCP/A2A protocols, grounded retrieval, internal platforms, and eval-driven delivery.

## What I Build
- Multi-agent coordination layers with worktree isolation, quality gates, and HMAC audit trails.
- Agent systems with clear tool contracts, approval boundaries, and observable state transitions.
- Retrieval and answer pipelines that cite, abstain, or escalate instead of improvising with confidence.
- Internal AI platforms that unify fragmented APIs, workflow automation, and operational visibility into a single operator surface.
- Spec-driven delivery loops that keep humans architectural while agents handle the mechanical bulk.

## Flagship Open-Source Work

### Bernstein — the open-source governance layer for AI agents
- Creator and maintainer of Bernstein (https://bernstein.run, Apache-2.0, https://github.com/sipyourdrink-ltd/bernstein).
- Open-source governance layer for AI agents, built on a deterministic multi-agent scheduler, distributed on PyPI.
- Coordinates 40+ AI coding-agent adapters in parallel — Claude Code, Codex, OpenAI Agents SDK, Cursor, Gemini CLI, Aider, Amp, and more.
- What demo-grade multi-agent tools do not have: worktree-isolated execution, janitor verification, quality gates, HMAC audit trail, cost-aware model routing.
- Supports MCP and A2A protocols, Cloudflare cloud runtime.

### HireEx — hosted job-discovery service
- Creator and maintainer of HireEx (https://hireex.ai, closed-source SaaS — no public repository).
- A quiet hosted service: drop a résumé once; every morning a ranked daily shortlist arrives with one-line pitch hints. Operated by Sip Your Drink Ltd.
- Pricing: Free (3 roles each morning), Plus $19/mo (10 roles + personalised pitch), Pro $49/mo (exclusive sources + follow-up reminders).

## Recent Systems Work

### Marketing operations CNS (case study)
- Sole engineer on a central operations platform built end-to-end in three months.
- Next.js + FastAPI + ClickHouse. Unifies real-time data from seven ad platforms and four revenue partners into a single AI-powered decision surface.
- Agentic AI advisor with three-tier tool-calling and MCP integration. Every answer grounded in live data via retrieval and citation gates. Anomaly detection, optimization recommendations, creative generation pipeline.
- Full product surface: kanban, notifications (in-app, push, email, Teams), RBAC, ChatGPT-style sharing. Became the team's primary daily operating tool.
- Public write-up: https://alexchernysh.com/blog/building-cns-marketing-operations

### Legal RAG — Agentic RAG Legal Challenge
- Placed 38th of 356 registered participants in the Agentic RAG Legal Challenge as team Tzur Labs, scoring 0.47968, with a grounded legal QA system — page-level citations, hybrid retrieval, evidence-first answers. Leaderboard: https://agentic-challenge.ai/leaderboard
- Public write-up: https://alexchernysh.com/blog/legal-answering-systems

### Independent AI consulting
- Privacy-first RAG for legal and compliance use cases.
- Customer-facing AI agents with traceability and operator controls.

## Core Stack

### AI & Agent Systems
- Multi-agent orchestration, RAG, evals, tool-calling architectures.
- MCP/A2A protocols, guardrails, cost-aware routing, audit trails.
- Grounding, retrieval and reranking, abstention and escalation patterns.

### Backend & Data
- Python, FastAPI, Next.js, React, TypeScript.
- PostgreSQL, ClickHouse, Redis, Qdrant.

### Infra & Delivery
- Docker, Ansible, GitHub Actions, Cloudflare, Hetzner, Prometheus, Kubernetes.
- Spec-driven delivery loops, worktree-isolated execution, eval-first rollouts.

## Working Style
- Architecture before ornament.
- Reliability before bravado.
- Fast thin slices over theatrical rewrites — but every slice carries grounding, evals, telemetry, and a rollback path.
- Measurable behavior: logs, evals, failure modes, audit trails.

## Typical Project Archetypes
- A multi-agent coordination layer for teams running many AI coding agents in parallel.
- A compliance-sensitive knowledge assistant for legal or policy-heavy work.
- An internal AI platform that unifies fragmented APIs, workflow automation, and operational visibility into a single operator surface.
- A customer-facing AI assistant with tool use, citations, and operator controls that can explain why an answer is trustworthy.

## Core Capabilities
- Multi-agent orchestration and coordination
- Agent orchestration with tool contracts and approval boundaries
- Grounded RAG, retrieval, reranking, page-level citations
- MCP / A2A protocol integration
- Eval design, regression checks, and release gates
- Cost-aware model routing and audit trails
- Observability, resilience, and rollback patterns
- FastAPI / Python delivery and TypeScript / Next.js product surfaces

## Preferred Questions
- What is Bernstein and what does it solve?
- What is HireEx?
- How do you coordinate multiple AI agents reliably?
- How do you keep a RAG system grounded?
- What would you audit first in an unreliable AI stack?
- How do you work with product and engineering teams?
- How do you trade off delivery speed against rigor?

## Canonical Public Answers

- What is Bernstein?
  Bernstein is the open-source governance layer for AI agents, at bernstein.run — Apache-2.0 on GitHub. It coordinates 40+ AI coding-agent adapters (Claude Code, Codex, OpenAI Agents SDK, Cursor, Gemini CLI and more) in parallel with worktree isolation, quality gates, janitor verification, HMAC audit trail, and cost-aware model routing. MCP/A2A supported. Designed for production reliability, not demo spectacle.

- What is HireEx?
  HireEx is my hosted job-discovery service at hireex.ai — closed-source SaaS, no public repository. Drop a résumé once; every morning a ranked daily shortlist arrives with one-line pitch hints. Free (3 roles/morning), Plus $19/mo (10 roles + personalised pitch), Pro $49/mo (exclusive sources + follow-up reminders). Operated by Sip Your Drink Ltd.

- How do I audit a shaky RAG stack?
  Start with retrieval quality and fallback behavior. Then check tool boundaries, eval coverage, citation/grounding gates, telemetry, rollback paths, and who actually owns the system once it is live.

- How do I think about agent boundaries?
  Explicit tool contracts, approval paths, and observable state transitions. Agents handle bounded delegated work and escalate edge cases instead of improvising authority. For multi-agent setups: isolation per task (worktrees), quality gates between hops, audit trail on every action.

- How do I trade off speed and reliability?
  Fast thin slices, but each slice ships grounding, evals, telemetry, and a rollback path. Speed only helps if the system stays legible when something breaks. Cost tracking and audit trails are first-class primitives.

- Where am I most useful in a team?
  Best where the model is no longer the interesting part: architecture, grounding, delivery shape, and making a live system trustworthy under real constraints. Senior / staff-level AI systems, platform engineering, or technical architecture.

- What is the marketing CNS case study?
  Sole-engineer build in three months: Next.js + FastAPI + ClickHouse central operations platform unifying seven ad platforms and four revenue partners. Agentic AI advisor with three-tier tool-calling and MCP integration, anomaly detection, optimization recommendations, creative generation pipeline. Full write-up at alexchernysh.com/blog/building-cns-marketing-operations.

- What was the legal RAG competition?
  The Agentic RAG Legal Challenge — placed 38th of 356 registered participants as team Tzur Labs, score 0.47968, with a grounded legal QA system. Page-level citations and hybrid retrieval. Leaderboard at agentic-challenge.ai/leaderboard, write-up at alexchernysh.com/blog/legal-answering-systems.


## Navigation Map

- https://alexchernysh.com — Homepage, positioning, contact (also hosts the bounded AI bot)
- https://alexchernysh.com/consulting — Consulting practice (multi-agent orchestration)
- https://alexchernysh.com/blog — Technical writing
- https://alexchernysh.com/bernstein — Bernstein product page
- https://alexchernysh.com/rightlayout — RightLayout macOS utility
- https://bernstein.run — Bernstein canonical site
- https://hireex.ai — HireEx canonical site (service closed 1 Aug 2026)
- https://github.com/chernistry — GitHub profile
- https://github.com/sipyourdrink-ltd/bernstein — Bernstein source
- https://pypi.org/project/bernstein/ — Bernstein PyPI
- https://bernstein.readthedocs.io — Bernstein docs


## Published Writing

Public writing archive for Alex Chernysh, focused on grounded AI systems, retrieval, evals, observability, and integration-heavy delivery.

## Published Notes

- [Your banned-word list expired in March 2024](https://alexchernysh.com/blog/edit-for-commitment) — Naming a tell is what kills it, so any list of AI words starts dying the day it ships. Underneath sits the failure no list touches: the draft commits to nothing. I built an editing skill for that one. (2026-08-12)
- [RightLayout: Shipping a Mac AI Tool, Then Letting Go](https://alexchernysh.com/blog/rightlayout-shipping-then-letting-go) — Why I trained a small CoreML model from scratch for a Mac keyboard-layout corrector, used it for months, and then open-sourced it instead of scaling it. (2026-05-08)
- [Forecasting Without Prophecy: a plain-text discipline](https://alexchernysh.com/blog/forecasting-without-prophecy) — Why I leave the future to astrology and reach for reference classes, premortems, and calibration logs instead. Disciplined uncertainty in plain text. (2026-05-02)
- [I Ran 12 AI Agents for 47 Hours. Here's What Survived.](https://alexchernysh.com/blog/bernstein-multi-agent-orchestration) — Open-source deterministic orchestrator for parallel CLI coding agents. Runs Claude Code, Codex CLI, Gemini CLI in parallel: zero coordination tokens, 40+ adapters, janitor verification, git worktree isolation. (2026-03-29)
- [Working Under Repeated Alarms](https://alexchernysh.com/blog/building-under-rockets) — A short note from Israel on what repeated alarms do to attention, engineering judgment, and team habits, and which working practices make interruption easier to absorb. (2026-03-11)
- [How to Build Legal Answering Systems That Can Be Trusted](https://alexchernysh.com/blog/legal-answering-systems) — A practical blueprint for legal QA, shaped in part by work around the Agentic RAG Legal Challenge: document identity, hybrid retrieval, structured answers, page-level grounding, telemetry, and evals. (2026-03-10)
- [LLM Product Safety Without Theater](https://alexchernysh.com/blog/llm-product-safety-without-theater) — A practical guide to LLM product safety: prompt injection, excessive agency, unsafe outputs, evals, and sober boundaries. (2026-03-09)
- [Interface Design for Serious Products](https://alexchernysh.com/blog/interface-design-serious-products) — A practical memo on calm authority, visible product care, restrained motion, and why trustworthy interfaces feel expensive. (2026-03-06)
- [Getting AI-Assisted Development to Green Without Breaking the Code](https://alexchernysh.com/blog/ai-assisted-development-green-state) — Repair loops, small diffs, test trust, and how to get CI back to green without trashing the codebase. (2026-03-04)
- [Building Agentic AI Systems That Hold Up](https://alexchernysh.com/blog/agentic-systems-best-practices) — Practical guidance on tool contracts, context engineering, evals, approvals, and telemetry. (2026-03-02)
- [Which Query Transformation Techniques Actually Help RAG?](https://alexchernysh.com/blog/query-transformation-for-rag) — HyDE, query rewrite, decomposition, step-back prompting, and fusion for RAG: which query transformation technique fixes which retrieval failure, and when the extra latency pays off. (2026-02-24)
- [Preventing Hallucinations in LLM Systems](https://alexchernysh.com/blog/llm-hallucination-prevention) — How to reduce hallucinations in LLM systems with better retrieval, abstention, verification, evals, and guardrails. (2026-02-18)
- [Most RAG Failures Start in the Documents](https://alexchernysh.com/blog/rag-document-preparation) — Chunking, titles, metadata, parent-child structure, reranking, and corpus QA for RAG systems. (2026-02-12)
- [Spec-Driven Development: the workflow I actually use](https://alexchernysh.com/blog/sdd-spec-driven-development) — How I use a lightweight spec-driven workflow in real projects, what SDDRush automates, and where Kotef fits if you want a stronger agent layer. (2026-02-06)
- [How to Run LLM Evals in Production](https://alexchernysh.com/blog/llm-evals-in-production) — LLM evals for continuous delivery: turn production failures into automated tests, grade traces with task-specific graders, and block bad releases with eval-driven gates. (2026-02-03)
- [Prompt Engineering: From Phrasing to Policy](https://alexchernysh.com/blog/prompt-engineering) — Prompt design now means response formats, examples, tools, and eval loops, not incantations. (2026-01-29)
- [BI Storytelling That Actually Moves Decisions](https://alexchernysh.com/blog/bi-storytelling) — How to make BI pages support decisions through narrative, visual hierarchy, and trust. (2026-01-14)
- [SYNAPSE: adaptive-control software engineering, prototyped](https://alexchernysh.com/blog/synapse_paper) — SYNAPSE was a 2025 framework for AI agents that adapt their own success criteria via MCDM. The deterministic-control-plane idea later shipped as Bernstein. (2025-07-15)