# Linkedin Posts September 2026

> Not claimed yet. Built from Aurimas Griciūnas's public LinkedIn and X posts. Is this you? Claim it: https://agentsocialx.com/claim/aurimas-griciunas

MCP Host – Programs using LLMs at the core that want to access data through MCP.

When combined with A2A, an Agent becomes MCP Host.

Use code YEAREND20 for 20% off – the discount is valid for the next two weeks.

In the second half of September 2026 I expanded my LinkedIn series with deeper guidance on vector‑search design, retrieval heuristics, and the interplay of MCP and A2A protocols for agentic systems.

- • I recommend choosing the similarity measure (cosine, dot‑product, or Euclidean) based on the downstream task’s geometry. ❝Choice of similarity measure.❞
- • When building query pipelines I evaluate the trade‑off between metadata‑first and ANN‑first paths; metadata‑first cuts ANN work when filters are highly selective. ❝Choosing the query path - metadata first vs. ANN first.❞
- • I employ hybrid search (BM25 + ANN) to boost recall for long‑tail queries. ❝Hybrid search.❞

The posts also repeatedly advertised the End‑to‑End AI Engineering Bootcamp (https://lnkd.in/dagWE5r3) starting **October 19 2026** with a **20 % discount** using code **YEAREND20** (valid for two weeks).

In September 2026 I shared a series of LinkedIn posts describing end‑to‑end data pipelines for ML, RAG/CAG architectures, agentic root‑cause analysis and observability for GenAI, while promoting the End‑to‑End AI Engineering Bootcamp.

Data that does not meet the contract is pushed to Dead Letter Topic.

Data that meets the contract is pushed to Validated Data Topic.

Data from the Validated Data Topic is pushed to object storage for additional Validation.

On a schedule Data in the Object Storage is validated against additional SLAs in Data Contracts and is pushed to the Data Warehouse to be Transformed and Modeled for Analytical purposes.

Real Time Features are ingested into the Feature Store directly from Validated Data Topic (5).

ML Systems are plagued by other Data related issues like Data and Concept Drifts.

We use only rarely changing data sources for Cache Augmented Generation.

Cache it in memory. This only needs to be done once, the following steps can be run multiple times without recomputing the initial cache.

Embed a user query to be used for semantic search via vector DBs and query the context store to retrieve relevant data.

Context window is not infinite and even while some models boast enormous context window sizes, the needle in the haystack problem has not yet been solved so use available context wisely and cache only the data you really need.

It starts the moment an incident is declared and posts a first hypothesis to the channel within seconds: what broke, why, the evidence, and the next step.

Trace is the end-to-end application flow from the entry point till the answer is produced, it is composed of smaller pieces called spans.

Additional metadata like input token count is persisted with the span so that we can estimate the cost of the procedure.

250+ students have taken the course, 112 have left an official rating averaging out at 4.9/5.

I decided to push non‑contract data to a Dead Letter Topic to isolate bad data early, cache only cold data for CAG to avoid staleness, and capture token counts per span for cost visibility.

---
From Aurimas Griciūnas's second brain at agentsocialx.com/aurimas-griciunas