Seth Juarez

The majority of companies interviewing developers are asking questions relevant 5-10 years ago - the tectonic plates have shifted, learn an AI-Native development workflow or become obsolete. Algorithms are secondary, structure and process are now king. 👑

3

10 Comments

Claude Skills - which improve how Claude performs specific tasks - were just announced on Oct 16th 2025. But, Travis Van Nimwegen shows how active the Claude community already is by publishing his "Awesome Claude Skill" Github repository, with a 3rd-party contribution list that augments by the day It aggregates not only the reference documentation and the skills officially published by Anthropic but also all the skills and companion tools already developed by users and developers . The MUST-VISIT spot if you (plan to) leverage Claude Skills ! Awesome Claude Skills repo: https://lnkd.in/eGZzDgyJ Claude Skills announcement: https://lnkd.in/eFJwG8QS #claude #anthropic #skills #llm #genai #agents

9

2 Comments

Using Angular? Still not transitioned to signal? You should! No memory leaks No noisy subscription code No .next, .value, .asObservable() clutter And no crying when refactoring a component Angular is moving towards singal-first reactivity, zoneless change detection, standalone components, small bundle sizes and faster startup. Adapting early will save you a lot of pain, doing the transition in smaller chunks. And best of all, it is so smooth and nice to work with! Really impressed with how Angular keeps position with their framework. In my opinion (which offcourse is a bit biased) it is the best tool if you are going to build something that can take some weight. If you want something fast and don't care about using community packages (with various degree of maintenance), you can use anything. But if you are going to live with the product for years, Angular is a great choice.

1

See how Forward Deployed Engineers have helped close the gap on implementing AI agents leading to a 800% spike in FDE job postings between Jan. and Sep. 2025, according to Indeed and the Financial Times:

3

As LLMs evolve from static models into dynamic components of real-world applications, they must do more than perform tasks; they must also adapt to context, personalize interactions, and continuously improve through feedback. This blog outlines a practical, modular approach to designing adaptive, production-ready LLM systems using: - Supervised Fine-Tuning (SFT) for core task competence - Reinforcement Fine-Tuning (RFT) for behavioral alignment - Context-aware routing for persona and adapter control - RAG for real-time knowledge injection - Telemetry loops for continuous learning and refinement It also includes a curated toolchain (LoRA, trl, vLLM, Langfuse, and more). https://lnkd.in/gBtjeg_X

17

2 Comments

How many different things can you do in an AI app built in Azure Foundry? Join us at Azure Cleveland to learn all about it: https://lnkd.in/gvKT_t_S

4

The recent discussion on PHP Architect's "Community Corner" with Larry Garfield about the PHP Framework Interop Group (FIG) truly resonated with me. While not always front-page news, the work of groups like FIG is foundational, enabling the seamless collaboration and component interchangeability we often take for granted in modern PHP development. This isn't just about cleaner code; it's about reducing friction, accelerating development cycles, and ensuring long-term maintainability for businesses relying on PHP applications. From a consultant's perspective, embracing these standards (like PSRs) is a non-negotiable best practice. It fosters an ecosystem where libraries and frameworks can coexist harmoniously, leading to more robust, scalable, and adaptable systems. In my Laravel projects, this commitment to interoperability extends beyond PHP itself – it informs how I design APIs for React frontends, structure reusable components in Flutter, or ensure data consistency with Python backends. It’s a core philosophy that drives robust, scalable, and future-proof solutions, minimizing technical debt and maximizing business value across the full stack. What "behind-the-scenes" standard or group has had the most significant, yet often overlooked, positive impact on your software development projects? #PHP #Laravel #SoftwareDevelopment #TechConsulting #Interoperability

1

This week I have been spending time on retrieval evaluation and optimization:  not just building retrievers, but benchmarking them systematically across recall, precision, latency, and cost. Notebook link for those interested in the benchmarking setup : https://lnkd.in/gctZMwgQ Here are the technical learnings that stood out: 1. Retrieval strategies encode different inductive biases: BM25 consistently outperformed dense embeddings for exact-phrase and parameter-heavy queries because it models lexical term frequency and document statistics directly. Embeddings performed better when semantic reformulation was required. This is not about “which is better,” but about understanding what each retriever is structurally capable of capturing. Choosing only one imposes a blind spot on the system. 2. Query expansion improves recall, but recall scales sublinearly with cost: Multi-query retrieval expanded the semantic search space and measurably improved context recall. However, each additional reformulation adds an LLM call and a retrieval pass. Recall gains plateaued while cost and latency increased linearly. Retrieval optimization is therefore a constrained optimization problem — maximize recall subject to latency and cost budgets. 3. Chunking architecture influences grounding more than embedding choice: Parent-document retrieval (small chunks for indexing, larger parents for response) improved grounding stability because it decouples retrieval granularity from context coherence. In contrast, semantic chunking struggled on repetitive FAQ-style corpora where similarity thresholds collapsed boundaries. Chunking is not preprocessing — it is part of retrieval system design. 4. Evaluation must be retrieval-specific, not generation-centric: Using synthetic gold datasets with retrieval-level metrics (context_precision, context_recall) surfaced failure modes that manual testing would not detect. Most RAG instability in practice is traceable to retrieval configuration decisions that are rarely measured rigorously. The broader takeaway: Retrieval performance is a systems problem. Precision, recall, latency, and cost form a coupled system. There is no universally optimal retriever, only configurations optimized for a corpus and constraint profile. Shoutout to AI Makerspace for the structured exploration. #RAG #RetrievalSystems #LLMEvaluation #SearchArchitecture #GenerativeAI #AIMakerspace

32

1 Comment