Skip to content
View mukeshrebari's full-sized avatar
🎯
Focusing
🎯
Focusing

Block or report mukeshrebari

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
mukeshrebari/README.md

Mukesh Rebari

Principal Architect & Co-founder — agentic AI systems and the distributed backbones they run on.

I design and ship production systems where LLMs do real work: multi-agent orchestration, retrieval, and finance/operations automation on top of event-driven, multi-tenant .NET and TypeScript services. The repositories below are clean-room reference implementations of patterns I run in production — each is small, runnable, tested, and documented as a build-along guide.


Selected work

🤖 Agentic AI & LLM systems

Project What it is Stack
mcp-domain-agents A custom MCP server exposing an internal service as LLM-callable tools, with a coordinator agent delegating to specialist sub-agents. .NET 10 · MCP + Semantic Kernel
finance-copilot A controller "brain" agent orchestrating three specialist service-agents (each with its own tools) via agent-as-tool delegation. .NET 10 · Microsoft Agent Framework
resilient-llm-gateway A provider-agnostic LLM gateway: ordered failover, circuit breaking, retries, cost-aware routing. TypeScript
self-improving-insights Deterministic, cited data insights that measurably improve from user + engineering feedback. TypeScript
document-rag Grounded RAG with citations that refuses when the corpus doesn't support an answer. TypeScript

🔗 Distributed systems

Project What it is Stack
saga-orchestration An event-driven saga with compensation for multi-service transactions — no distributed lock. .NET 10 · MassTransit

📊 Evaluation

Project What it is Stack
ollama-model-bench A reproducible LLM benchmark — 902 judged responses across 22 models — with an LLM-as-judge rubric. Node

How I think about systems

  • Determinism first. Compute exact facts in code; let the model narrate, not calculate.
  • Least privilege. Narrow, typed, read-only tools — the blast radius is the tool surface, not the database.
  • Design for failure. Compensation over two-phase commit; failover and circuit breakers over single-provider hope.
  • Refuse before you guess. Grounded answers with citations; a clean refusal when the evidence is thin.
  • Measure before you scale. Evals and feedback loops that prove a change helped.

Stack

.NET / C# · TypeScript · event-driven (MassTransit, Azure Service Bus) · Semantic Kernel · Microsoft Agent Framework · Model Context Protocol · RAG · PostgreSQL · Redis


Each repository above ships a README (problem, architecture, design decisions, trade-offs) and a step-by-step GUIDE, on synthetic/neutral domains — the reusable patterns, not any employer's code.

Pinned Loading

  1. ollama-model-bench ollama-model-bench Public

    Reproducible LLM benchmark across coding, reasoning, and knowledge tasks with an LLM-as-judge rubric

    JavaScript 1

  2. document-rag document-rag Public

    Grounded retrieval-augmented generation with citations and refusal when the documents don't support an answer

    TypeScript 1

  3. mcp-domain-agents mcp-domain-agents Public

    Custom MCP server exposing a domain service as LLM-callable tools, with a coordinator agent delegating to specialist sub-agents

    C# 1

  4. resilient-llm-gateway resilient-llm-gateway Public

    Provider-agnostic LLM gateway with ordered failover, circuit breaking, retries, and cost-aware routing

    TypeScript 1

  5. saga-orchestration saga-orchestration Public

    Event-driven saga orchestration with compensation for multi-service transactions

    C# 1

  6. self-improving-insights self-improving-insights Public

    Deterministic dataset insights narrated by an LLM, improved over time via a user and engineering feedback loop

    TypeScript 1