Looking to connect with people working in AI engineering · backend systems · LLM products what I'm building: → agentic RAG systems from scratch → GraphRAG for codebase intelligence → local inference pipelines what I want to talk about: → retrieval architecture → prompt engineering that's actually measured → building AI products without framework dependency drop what you're working on below always down to exchange ideas 👇 #AIEngineering #Backend #buildinpublic #LLMs

Here is the CLI working; graphRAG is really efficient for code-base relational chunking and retrieval. Check this out: github.com/reloaddd/Agent-PN…

Built something that's been bothering me for months Every code RAG I've seen treats source code like an essay: chunk at 500 chars, embed, retrieve, hallucinate Functions split in half, inheritance chains invisible test files returned instead of core logic so I built Agent-PNSV 👇 Instead of character limits → Tree-sitter AST parsing chunks at actual compiler boundaries function definitions, class definitions, complete and intact Instead of vector search only → dependency graph CALLS, INHERITS_FROM, INCLUDES when you ask about a function, it fetches the parent class, the helper functions, the full context Point it at any GitHub repo it clones, parses, indexes, maps the graph ask an engineering question local Ollama answers with actual structural context fully local no cloud APIs no data leaving your machine two interfaces: → Streamlit dashboard with AST payload inspector → CLI for terminal people code is a graph your RAG should know that github: github.com/reloaddd/Agent-PN… #buildinpublic #AIEngineering #RAG #CodeIntelligence

looking to connect with people building in the AI space founders · AI engineers · researchers · builders specifically interested in: → anyone working on RAG systems or agents → building LLM products from scratch → thinking about evals and production AI drop what you're building in the comments let's connect 👋 #AIEngineering

I can't find any good excuse not to grind LeetCode

Integrated Prompt engineering and eval into Agent-Kiro 🖇️github.com/reloaddd/Agent-Ki… Not the basics — actual production techniques What I implemented: → Versioned prompt system every prompt change is tracked eval runs on each version no more "it feels better" guesses → Query rewriting "Summarize document?" becomes a specific retrieval query vague questions now find the right chunks → Chain of thought auto-detection complex questions trigger step-by-step reasoning simple questions stay fast → Temperature control 0.0 for routing — deterministic, no randomness 0.1 for answers — precise but natural The interesting part: adding query rewriting dropped eval from 90% to 60% Rewrites were too aggressive lost document context Router sent everything to web instead of personal content --> rewrite the query optimizer --> 80% eval LESSON: Prompt engineering without evals is just vibes Measure everything!! #buildinpublic #AIEngineeringg

Its been a year I have solved DSA problem, I think i should get back to it.

Started AI engineering with ML knowledge a few weeks ago decided to build instead of watch tutorials here's what happened 🧵 —— Week 1: Built a RAG pipeline from scratch no LangChain, no shortcuts just Python, embeddings, vector search, and a lot of googling vision OCR to read handwritten PDFs chunking strategies, semantic retrieval LLM-as-judge evals — hit 90% accuracy —— Week 2: Made it agentic agent that decides — search the PDF or search the web? tool routing, conversation memory, per-user sessions deployed it as a full stack app with FastAPI shipped Kiro — chat with any PDF using AI 📁 github.com/reloaddd/Agent-Ki… —— Week 3 and beyond: Going deeper prompt engineering → context engineering semantic caching → observability with Langfuse reranking → HyDE → query rewriting streaming responses → multi-document support YouTube URL ingestion → beyond just PDFs flashcards quiz mode → user auth persistence Goal: Turn Kiro into a proper AI knowledge base where you can actually learn anything from it —— building everything in public documenting every failure and fix if you're on this path too, let's connect 👋 #buildinpublic #AIEngineering #LLMs #RAG

A must-read for workflows and agents towardsdatascience.com/a-dev…

After grinding for 30 hours over AI engineering I have built- -- A workflow with no framework, just pure python with 3 tools. -- Converted the same workflow into agent-loop architecture I am not sure if I need Harness engineers at the moment but I am feeling the need of it, so that I can apply harness engineering into this same agent-loop to understand performance improvisation or harnessing. Any suggestions or resource to learn harness engineering? Or what should I do next if not harness engineering.

As an Indian, do you know that Modi has an entire 24/7 grooming and image-management team hired from abroad, including stylists and beauticians from Italy and the West.? The monthly expense reportedly crosses ₹10 crore. Is such extravagant spending on optics justified?

🚨🇮🇱🇮🇷 This is the moment multiple Iranian missiles landed directly on Haifa, while the so-called "Iron Dome" failed to intercept them…