Weekly Digest
Issue #01Week of 8–14 Nov 2026Sample issue

This week in tech, explained for you.

Sample issue. These are evergreen explainers that show every feature of the site, not this week's news. The first real Sunday run replaces them.

5 stories35 min of deep divesPublished Sun 15 Nov 2026, 7:30 pm IST

The week in 60 seconds

  1. GRPO trains reasoning models with RL but drops PPO's critic: each answer is scored against a group of answers to the same prompt. Read: GRPO, explained: reinforcement learning without a critic →
  2. Durable execution records every step an agent takes so a crash means resume, not start over. Read: Durable execution: why long-running agents need a workflow engine →
  3. Indirect prompt injection hides instructions in content an agent reads (pages, emails, tool results) rather than in the user's message. Read: Prompt injection through tool outputs →
  4. Hash-based signatures such as SLH-DSA rely only on hash functions, which makes them a conservative post-quantum choice for chains like Ethereum. Read: Hash-based signatures and Ethereum's post-quantum plans →
  5. Hybrid search in Postgres combines full-text and pgvector similarity, then fuses the two rankings into one. Read: Postgres + pgvector hybrid search as agent memory →
ResearchAgents

GRPO, explained: reinforcement learning without a critic

Group Relative Policy Optimization samples several answers per prompt, scores them, and pushes the model towards the ones that beat the group average. No value network is needed, which saves a model's worth of memory and made large-scale RL on reasoning practical for DeepSeek.

Why you ·Your agent work at HivePro and Susanoo lives on tool-use and reasoning quality. GRPO is the cheapest credible way to RL-tune a small model on a verifiable task, such as a JanusGuard triage policy scored against labelled CIC-IDS2017 flows.

14 minintermediateBuilds on PPO, policy gradients

Also this week

AgentsSystems
6 min

Durable execution: why long-running agents need a workflow engine

A durable execution engine persists the result of every side-effecting step as an event. If the worker crashes, another worker replays the history, skips the steps already done, and carries on from where it stopped. For agents that call slow, paid LLM APIs and tools, that turns crashes from expensive into boring.

Why you ·You already run Temporal next to LangGraph at HivePro. Knowing exactly what replay does (and why workflow code must be deterministic) is the difference between a resilient agent platform and mysterious non-determinism errors in production.

Open the deep dive: Durable execution: why long-running agents need a workflow engine
SecurityAgents
5 min

Prompt injection through tool outputs

When an agent pulls in a web page, an email or a tool result, any text inside it lands in the same context window as your instructions. An attacker who controls that content can try to steer the agent. The defences are architectural: least privilege, separating data from instructions, and human confirmation for risky actions.

Why you ·Every tool you add to the HivePro agent platform or Susanoo is a new input channel an attacker might control. JanusGuard also needs this threat model if it ever reads attacker-supplied logs or payloads.

Open the deep dive: Prompt injection through tool outputs
Web3
5 min

Hash-based signatures and Ethereum's post-quantum plans

Ethereum accounts are secured by elliptic-curve signatures, which a large enough quantum computer could break. Hash-based signatures rest only on the security of a hash function. NIST standardised one, SLH-DSA, in FIPS 205; the trade-off is much larger signatures.

Why you ·Post-quantum migration is a live research area where a careful survey or benchmark could become a paper, and it is the kind of cryptography depth that makes your Web3 work stand out at GDG talks.

Open the deep dive: Hash-based signatures and Ethereum's post-quantum plans
SystemsAgents
5 min

Postgres + pgvector hybrid search as agent memory

Vector search finds passages that mean the same thing; keyword search finds exact names, error codes and IDs. Postgres can do both: built-in full-text search and the pgvector extension. Run both queries and merge the ranked lists, for example with reciprocal rank fusion, to get memory that handles both kinds of question.

Why you ·Hybrid RAG is already part of your HivePro stack, and doing it inside Postgres builds directly on the Postgres internals you're learning. It's also a cheap memory layer for Susanoo before you need a dedicated vector database.

Open the deep dive: Postgres + pgvector hybrid search as agent memory
Try this weekend

GRPO fine-tune a 0.5B model on grade-school maths

About two hours on one free GPU: a small instruct model, a correctness reward and Hugging Face TRL's GRPOTrainer.

Steps in the deep dive
Tune next week's issue

What should change?

Each button opens a pre-filled GitHub issue. The Sunday run reads them and adjusts.

More agentsMore securityDeeper mathsLess cryptoShorterLonger