I build AI that actually ships.

AI engineer and full-stack developer. I turn large language models into products people use, from offline inference on a laptop to agent servers and the systems a business runs on every day.

ask-rahul local · idle

Scripted mode: answers written by me, streamed like a local model. Nothing leaves your browser.

1,000+ npm downloads since launch live
27 public repositories live
26 technical articles published live
~60% faster query turnaround from LLM agents at InferWorks

01 — Selected work

Open-source tools people install.

01 / Local AI

Unmuzzled

Private, offline AI chat that runs entirely on your own computer. One command, no account, no API key, no cloud.

  • Runs open-weight GGUF models through llama.cpp, with CUDA, Vulkan or Metal acceleration and a CPU fallback.
  • Streaming Markdown, a live reasoning view, edit-and-regenerate and searchable history, all stored locally.
  • 500+ downloads in its first week on npm.

TypeScriptNode.jsllama.cppGGUF

$ npx unmuzzled
✓ model ready · running on this machine
→ open http://localhost:3000
npm downloads since launch live—

02 / LLM cost engineering

ClaudeCut

A drop-in local proxy that cuts Claude API and Claude Code bills by 40–70% without touching your code.

  • Places prompt-cache breakpoints correctly, so repeated context is billed at the cached rate instead of full price.
  • Switches models automatically by task and shows spend on a live cost dashboard.
  • 500+ npm downloads in two weeks.

PythonFastAPIAnthropic APIPrompt caching

Monthly Claude spend, same workload

Before
100%
After
30–60%

Savings range from the project's published benchmarks.

npm downloads since launch live—

03 / AI agents

LinkedIn Guardrail MCP

A Model Context Protocol server that lets AI agents work on LinkedIn without putting the account at risk.

  • Works with Claude Desktop, Cursor, Windsurf and any MCP client: search jobs, read profiles, post and message.
  • Safe by design: a normal browser login instead of copied cookies, human-paced actions and hard daily quotas.

TypeScriptMCP · JSON-RPCPlaywright

  1. AI agentClaude · Cursor · Windsurf
  2. Guardrail MCPquotas · pacing · safe login
  3. LinkedInjobs · profiles · posts · inbox
Profile views today
Messages today
Delay between actions30–90 s

04 / Retrieval

NeuralDocs

Ask questions across large PDFs and get answers that cite the exact source passage, in under a second.

  • FAISS vector search over document chunks, with Llama 3 on Groq for fast generation.
  • Every answer links back to its source, so users can check it instead of trusting it blindly.

PythonFlaskLangChainFAISSLlama 3

What is the notice period for contract termination?

Either party may terminate with 30 days' written notice[1], or immediately for material breach[2].

  • [1] agreement.pdf · p. 14 · §9.1
  • [2] agreement.pdf · p. 14 · §9.3

answered in 0.6 s

02 — Experience

Where I've shipped for real users.

Jul 2026 – Present

AI Engineer & Full-Stack Developer · Aryan News Agency

Sole engineer. I own the agency's AI and software end to end, from roadmap and architecture to deployment and maintenance.

  • Built a dynamic, database-driven website with an LLM content pipeline that writes summaries, tags and SEO metadata, so staff publish without a developer.
  • Built the in-house accounting app from scratch: customer accounts, billing and invoicing, payments, outstanding dues and reports, with role-based access.
  • Added an AI assistant that answers plain-English questions about the books over read-only queries and drafts payment reminders.
Nov 2025 – Feb 2026

Founder's Intern & Full-Stack Developer · InferWorks, Bengaluru

  • Built LLM chatbots and agent workflows with LangChain and Llama 3, cutting query turnaround by about 60%.
  • Integrated Zoho CRM with backend datastores, automating lead capture and sync for 5,000+ records.
  • Deployed production apps on AWS with CI/CD through GitHub Actions, and led technical SEO across company sites.
2023 – 2026

Bachelor of Computer Applications · CHRIST (Deemed to be University)

Toolbox, as an embedding space

Hover a skill to find its nearest neighbours.

AI & LLMs
LLM agents, MCP, RAG, LangChain, FAISS, embeddings, prompt engineering
Models & APIs
Claude API, OpenAI API, Llama 3 on Groq, llama.cpp, OpenCV
Back end
Python, FastAPI, Flask, Node.js, PostgreSQL, SQLite
Front end
TypeScript, Next.js, React 19, Vite, Tailwind CSS
Ship it
AWS, Vercel, Docker, GitHub Actions, npm publishing

03 — Writing

Notes from the build log.

All articles on dev.to ↗
  1. I built a private, offline AI chat you can run with one command5 min
  2. How I built a safe LinkedIn MCP server for AI agents2 min
  3. OKF vs RAG: why RAG alone is failing enterprise AI agents3 min
  4. How I cut Claude API and Claude Code bills by 70%3 min
  5. Passion Pilot: AI-powered learning journeys3 min

04 — Recognition

A few milestones.

UN Millennium Fellow2025–2026 · Global leadership program convened by United Nations Academic Impact.
National Science Day, 1st place2025 · Smart Campus Navigation System.
National Science Day, 2nd place2024 · Biometric Verification.

05 — Contact

Have something that needs to ship?

I'm open to AI engineering and full-stack roles, and to the occasional interesting build. The quickest way to reach me is email.

Send an email

Prefer the terminal? npx rahulagarwal18