skip to content
rajneeshsoftware developer · coding since 2014

$ ls ~/writing

writing

notes from building software since 2014: how a server took the traffic, which database to pick, pipelines that deploy themselves, and the ai i build.

15 posts · 2024–2026aiscalingdatabaseswebdevops

latestai · 3 min read

The six layers I keep building into every AI system

Health data, client email, video, music. Different products, same skeleton. Here are the six layers that kept showing up in every AI system I built this year, and why each one exists.

read it

$ ls -t ~/writing | tail -n +214 more

  1. ai

    From a scanned PDF to clean JSON an AI can trust

    Medical reports, invoices, lab sheets. They all arrive as messy PDFs. Here is the pipeline I use to turn them into structured JSON, and why I never send the raw PDF straight to an LLM.

    2 min read
  2. ai

    Giving an AI long-term memory with vectors and RAG

    An AI that forgets every conversation is a toy. Here is how I give AI systems a real memory, from turning messages into vectors to matching them with the right client and answering from that memory.

    3 min read
  3. ai

    Privacy before inference: strip personal data before the AI sees it

    If your AI reads health records or client email, the safest personal data is the data the model never receives. Here is how I remove PII before inference, and put it back after.

    3 min read
  4. ai

    Never depend on one AI provider: a small layer that switches models

    Prices change, models change, free tiers disappear. Here is the small provider layer I put in every AI product, so switching between cloud APIs and a local Ollama model is a config change, not a rewrite.

    2 min read
  5. ai

    Teaching AI agents to fill a whole form in one call

    Browser agents fill forms like a nervous intern: screenshot, click, type, screenshot again. Here is the approach I built to read an app's code once and give the agent one skill per page, so a form takes one call instead of fifty.

    3 min read
  6. ai

    Testing AI agents before you ship them

    Normal tests check that 2 + 2 = 4. AI agents can answer differently every time. Here is how I evaluate agents with test sets, head-to-head comparisons, multi-turn runs and a scorecard, before any of them reach a user.

    3 min read
  7. ai

    Turning a long video into shorts with a local AI pipeline

    Paste a link, get vertical clips with captions, titles and tags. Here is how a local, low-cost shorts pipeline works step by step - transcription with Whisper, picking the best moments, face-tracked 9:16 crops and burned-in captions.

    3 min read
  8. ai

    Spotting risk over time: let code do the maths and AI do the explaining

    One lab report tells you little. Ten reports over two years tell a story. Here is how I build systems that track values over time, flag sudden changes and rising risk early, and explain them in plain language.

    3 min read
  9. ai

    Making your website readable by AI agents with llms.txt

    People now ask ChatGPT and Claude about you before they visit your site. Here is how I made mine readable by AI - a generated llms.txt profile, access for AI agents only, an "ask your AI" button, and one cache bug that taught me a lot.

    3 min read
  10. ai

    "I'll help you out" vs "I'll clarify this for you." An engine that knows the difference.

    AI systems are everywhere now. Clients also use AI tools to write emails. We use AI while researching, writing replies, preparing documents, planning work, and many other things. A lot of our communication now comes through AI. And that is what I call human involvement. I love this. But we also need to save the trees. All right, back to my stuff.

    7 min read
  11. scaling

    Scaled Server for High Traffic App around 25k hits daily

    Distributed Systems and Database Scaling

    4 min read
  12. databases

    All about RDBMS (SQL) | Benefits and limitations of SQL & NoSQL

    Many modern applications often use both SQL and NoSQL databases together to get the best of both worlds!

    5 min read
  13. web

    Share Anything with native share options in Html & JavaScript

    The Web Share API is a valuable feature that can enhance your website's functionality and user experience. It allows you to share text, URLs, or files with other apps or services using the…

    2 min read
  14. devops

    Leveraging Jenkins with Private GitHub and EC2 for Seamless Development Workflow

    As developers, we often have to go through many steps to build, test and deploy our project. But what if we could automate these tasks and streamline the process? GitHub already offers some…

    6 min read