Jul 2026Local LLMs
Meet Ornith: The Agentic Coding Model That Runs Entirely on Your Laptop
No API keys, no cloud bill, no code leaving your machine — just a 5 GB model on a fanless MacBook Air.
Sajith K
Chief Technology Officer, Ideenkreise Tech · Ernakulam, Kerala, India
I make AI actually useful.
Former assistant professor, now CTO. I work on agentic coding and on getting AI agents into everyday work — then I write down what actually held up, and what didn't.
- Agentic coding
- Everyday AI workflows
- Local LLMs
- LangGraph / RAG
Articles
25 published on MediumPractical writing on working efficiently with AI — agentic coding, local models, and putting agents to work in everyday life. What held up, and what didn't.
Jun 2026Agentic coding
Engineering With AI, Not Vibe Coding
Mar 2026Agentic coding
The Future Belongs to Thinkers, Not Coders — and These Two Tools Prove It
- GSD-2: The AI Coding Agent That Actually Controls Its Own Context Window
The original Get Shit Done went viral. Now it's being rebuilt from scratch — and the difference matters.
- From Concept to Production in 26 Days: A Case Study in AI-Augmented Development
Building an enterprise hiring platform with Windsurf Cascade, Claude Opus and Google Stitch — in an accelerated timeframe.
- From Linear Chains to Cyclic Graphs: Supercharging Your Code Interpreter with LangGraph
Advancing a secure Docker sandbox with LangGraph's cyclic-graph architecture for richer LLM capabilities.
- Build Your Own “Code Interpreter”: A Secure Docker Sandbox for LLM Agents
LLMs are brilliant at language but stumble on precise computation. Here's a sandbox that fixes that.
- Deploy Hugging Face SLMs on CPU with llama.cpp + Nginx Proxy
Running small language models efficiently on CPU infrastructure using llama.cpp — no GPU required.
- Deploy Hugging Face SLMs on CPU with Ollama + Nginx Proxy
A step-by-step guide to running quantized LLMs locally on an affordable VPS without a GPU.
- Using a Custom Chat Model Class in LangChain to Integrate Hugging Face Models with LangGraph
Creating custom chat-model classes in LangChain to connect Hugging Face models into LangGraph workflows.
- Agno vs LangGraph: Which AI Framework Wins on Speed?
Benchmarking two popular agent frameworks head-to-head — and separating the speed claims from reality.
- Simplify LLM Integration: Convert FastAPI Endpoints to an MCP Server with FastMCP
Turn traditional FastAPI endpoints into an LLM-friendly MCP server with almost no extra code.
- Unlocking the Power of Transformer Models with Hugging Face's Pipeline
A hands-on intro to Hugging Face's pipeline API for running transformer models on real NLP tasks.
- Create and Deploy Your Own Lightweight LLM API with Gemma 3 270M and LitServe
Build and ship a tiny, fast LLM API with Gemma 3 270M and LitServe — no heavy infrastructure needed.
- Introducing FastMCP v2: The Pythonic Way to Build Secure MCP Servers and Clients
Give LLMs seamless, secure access to external data and tools with the Pythonic FastMCP v2.
- Building a Real-Time Speech-Enabled LangGraph Agent with LiveKit
Wire a LangGraph agent to LiveKit for real-time, two-way voice interaction.
- CopilotKit (AG-UI): Integrating LangGraph with Next.js and FastAPI
Put a LangGraph agent behind a real UI — wiring CopilotKit's AG-UI to a Next.js + FastAPI stack.
- Creating an MCP Server and Integrating with LangGraph
Build an MCP server from scratch and plug it into a LangGraph agent as a standardized tool source.
- Using PostgreSQL with LangGraph for State Management and Vector Storage
Back a stateful LangGraph app with PostgreSQL for durable state and vector storage.
- Vibe Coding a Space Shooter Game with Windsurf and Claude 3.5 Sonnet
The build story behind the playable Space Shooter experiment — coding a browser game with Windsurf and Claude.
- Building a Chatbot with Long-Term Memory Using LangGraph
Give a chatbot durable long-term memory so it remembers across sessions, using LangGraph.
- Understanding Short-Term Memory in LangGraph: A Hands-On Guide
Chatbots fall apart when they forget what you just said. Here's how short-term memory works in LangGraph.
- Deploying LangGraph with FastAPI: A Step-by-Step Tutorial
Ship a LangGraph chatbot as a FastAPI service, powered by LangChain's Groq integration.
- Implementing a Prompt Generator Using LangChain, LangGraph, and Groq
Build a human-in-the-loop prompt generator that turns user requirements into polished prompts.
- Harnessing the Power of LLM Agents in Software Development
Where it all started — exploring LLMs, LangChain and LangGraph to harness agents for real dev work.
Journey
From academia to shipping productsExperience
Chief Technology Officer
2021 — PresentIdeenkreise Tech Pvt Ltd · Ernakulam, Kerala, India
- Lead the development of innovative tech solutions end to end.
- Collaborate with a team of engineers to push technological boundaries.
Instructor · AI Foundations
2026 — PresentHours Back · Online · taught in Malayalam
- A six-week applied AI course for working professionals — accountants, admin, sales and teachers.
- Students build AI workflows for their own jobs, and measure the hours they get back.
Assistant Professor
2014 — 2020University College of Engineering, Muttom · Thodupuzha, Kerala, India
- Taught and mentored students in Computer Science and Engineering.
- Built a hands-on learning environment to prepare students for industry.
Education
Master of Technology (MTech)
2011 — 2014Computer Science and Engineering
Viswajyothi College of Engineering, Muvattupuzha (MG University)
CGPA: 8.33 / 10
Bachelor of Technology (BTech)
2007 — 2011Computer Science and Engineering
Matha College of Engineering, N. Paravur (MG University)
Percentage: 74.4%
Research & recognition
Publication
Abandoned or Removed Objects Detection from Surveillance Video Using Codebook
International Journal of Engineering Research & Technology (IJERT) · May 2013
A system for detecting abandoned and removed objects in surveillance video from stationary cameras, addressing occlusion and rapid lighting changes through foreground analysis.
Read the paper →Recognition
- Microsoft Technology Associate — Web Development Fundamentals
- Score: 92%
- Graduate Aptitude Test in Engineering (GATE) 2011
- 95th percentile
- Industrial Training — Advanced Computer Networking
- RTTC, Trivandrum, Kerala, India
Experiments
Built with AI. Playable, right now.
AI-assisted build
Space Shooter
A fast arcade space-combat game — enemy waves, power-ups and particle FX at a steady 60 FPS. Built and iterated on with AI pair-programming, then shipped as a static site.
Built by a local AI agent
Brick Breaker
Written end-to-end by Ornith-9B — an agentic coding model running fully offline on a fanless MacBook Air: one prompt to a complete game with paddle physics, particles, lives and levels. Later refined with GLM 5.2 to be fully mobile-friendly (touch controls + responsive canvas).