Skip to content
View Raghu23-dev's full-sized avatar
🎯
Focusing
🎯
Focusing

Block or report Raghu23-dev

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
Raghu23-dev/README.md
Raghuram P — GenAI Full-Stack Engineer
What I do
Profile views Followers LinkedIn Email Open to work

6 systems 3K+ users

🚀 About Me

I build the AI tools other engineers build with.

  • 🧠 GenAI Full-Stack Engineer — I design and ship AI-native enterprise platforms end-to-end: from multi-agent LLM orchestration and RAG retrieval down to Angular frontends, async FastAPI services, and the real-time streaming backbones that hold them together. My work reaches thousands of developers, designers, and product teams daily.
  • 🤖 Co-built & production-hardened an AI process-orchestration engine — multi-step LLM workflows with human-in-the-loop approval gates, governed multi-provider model routing, Kafka multi-pod streaming, conversation compaction, and per-run cost tracking. (2K+ daily users.)
  • 🧑‍💻 Built the backend for an AI coding plugin for VS Code (a Cursor / Claude-Code-class tool) — re-architected a single-agent prototype into a multi-agent orchestration system (Main-Agent + sub-agents on CrewAI, 30+ tools, skills & command frameworks, conversational memory, streaming). (3K+ daily users across 5+ enterprise environments.)
  • 🔎 Contributed to a hybrid code-retrieval RAG pipeline — Tree-sitter parsing + SQLite FTS5 BM25 + sqlite-vec + embeddings + Cohere rerank + RRF fusion + an 8-intent query planner. Owned the ingestion path: resumable multi-phase indexing, SSE progress streaming, two-tier admission control.
  • Designed an SSE + Redis Streams real-time backbone — replaced client polling with a decoupled pub/sub transport: per-session connection deduplication, exponential backoff, Last-Event-ID checkpointing, and history backfill on zero-replay reconnect.
  • 🎨 Co-built a GenAI wireframe generator (briefs/sketches/prompts → UI artifacts) and a prompt-to-React code-gen system; led a React → Angular re-architecture; and shipped a 126-component enterprise design system as a private NPM package.
  • 🌱 Currently going deeper on LLMOps, evaluation harnesses, and distributed-systems design.
  • 💬 Ask me about multi-agent orchestration, production RAG, and real-time streaming at scale.
  • 📫 Reach me at raghu.builds@gmail.com · 💼 LinkedIn

🧩 Systems I've Shipped

Built for enterprise clients across hardware, banking & fintech domains. Source is proprietary — company and product names withheld; architecture, tech, and scale are summarized below.

🤖 AI Coding & Agent Infrastructure

System What it is & tech Scale
AI Coding Assistant (VS Code plugin) In-editor AI pair-programmer grounded in your codebase — multi-agent orchestration (Main-Agent + sub-agents), 30+ tools, skills & command frameworks, conversational memory, compaction, human-in-the-loop execution, real-time streaming.
Python · FastAPI · CrewAI · ReAct · SSE · AWS Bedrock · Azure OpenAI
3K+
daily
AI Process-Orchestration Engine Runs multi-step LLM workflows as pluggable process-packs — HITL approval gates, governed multi-provider routing, Kafka multi-pod streaming, compaction, per-run cost tracking.
Python · FastAPI · CrewAI · Apache Kafka · SSE · Redis · OpenTelemetry
2K+
daily
Hybrid Code-RAG Pipeline (contributor; owned ingestion) Codebase indexing + hybrid retrieval grounding the agents — lexical + semantic + structural signals fused with RRF and an 8-intent query planner.
Tree-sitter · SQLite FTS5 BM25 · sqlite-vec · Azure OpenAI · Cohere Rerank · RRF
powers
the agents

🎨 AI App-Generation & Platforms

System What it is & tech Impact
AI App-Generation Platform Prompt-to-app, image-to-app & wireframe generation — designed the SSE + Redis Streams real-time backbone; re-architected the frontend React → Angular.
Angular · RxJS · Signals · FastAPI · Redis Streams · SSE
prompt →
running app
AI Product-Planning Platform Multi-step AI-assisted planning flow (Brainstormer → Persona → SWOT → Roadmap → Summary) — async services on Postgres.
Angular · FastAPI · PostgreSQL · asyncpg · AWS RDS IAM · Alembic · Redis
end-to-end
owner
Enterprise Angular Design System 90+ standalone components — OnPush change detection, tree-shaking, token-driven theming, GSAP animation infra; published as a private NPM package.
Angular · TypeScript · SCSS · GSAP · ng-packagr
90+
components

🛠️ Tech Stack


Python

TypeScript

JavaScript

FastAPI

Angular

React

Next.js

Tailwind

SCSS

gRPC

WebSocket

Nginx

PostgreSQL

Redis

Kafka

SQLite

MongoDB

Docker

K8s

AWS

Azure

Prometheus

Grafana

Playwright

Linux

Git

GitHub

VS Code
Every icon above maps to a line in my verified skills inventory — ask me to walk you through any of them.

🧠 GenAI / LLM   CrewAI Multi-Agent RAG ReAct HITL AWS Bedrock Azure OpenAI Cohere Tree-sitter Vector DBs

⚙️ Backend · Real-Time · Observability   Pydantic SSE Redis Streams Event-Driven OpenTelemetry

🎨 Frontend extras   RxJS Signals Monaco GSAP


🐍 Contribution Graph

Contribution snake eating my GitHub contribution graph
Sparse by construction, not by inactivity — most of the work above was committed under a work-domain email GitHub cannot link to this account. Attribution is fixed going forward; this graph only reflects what it can see.

⚡ Philosophy

Principles

I work best where product UX, distributed systems, and AI infrastructure meet — building tools that take people from an idea to working software, faster.


🤝 Connect

LinkedIn Email GitHub

footer

Pinned Loading

  1. fusegrid fusegrid Public

    A fuse for LLM spend. Four conventional enforcement patterns fail open at 95-100% over ceiling; this holds all four.

    Python

  2. mcpgauntlet mcpgauntlet Public

    Conformance and load harness for MCP servers on spec 2026-07-28. The first server it audited was my own: 6 MUST violations of 8.

    Python

  3. trigsight trigsight Public

    Portfolio with an AI assistant that cannot make a claim it can't prove — every citation is bound at build time to the exact supporting sentence, or the build fails.

    TypeScript

  4. onewayglass onewayglass Public

    Permission-aware retrieval: the result count leaks what it hid. 17 documents revealed becomes 0 — and a relevance channel that restores the full leak, published.

    Python