A DotNet developer’s note on Technology Radar V32

 

As a new Technology Radar release with many interesting techs, a few drew my attention. I picked a few looked interesting for dotnet developer.

Surprised by so many tools, frameworks around LLM application. It is definitely exploding.

Techniques

Adopt

  • Data product thinking: treat data as a product with its own lifecycle, quality standards and focus on meeting consumer needs. read more
  • Fuzz testing: test a system with a invalid inputs. This skill is more meaningful in the area of vibe coding. e.g. invalid input should result in 4xx error, not causing 5xx error.
  • Software Bills of Materials (SBOM): a comprehensive, machine-readable inventory of all components, library, and dependencies within a software product. Tools like Trivy and Snyk can help to find and fix the vulnerability of the software.
  • Threat modeling: a set of techniques for identifying and classifying potential threats. It’s more meaningful in the age of Gen-AI.

Trial

  • API request collection as API product artifact: This sounds like a good idea if API is public or consumers base is large.
  • Architecture advice process: :Let’s the team decide architecture by getting advices from those are affected and those having expertise rather than top-down decision.
  • GraphRAG: 1) chunk documents and create a knowledge graph using LLM 2) retrieve relevant chunks via embeddings and edges in the knowledge graph.
  • Just-in-time privileged access management: Provide least privilege with time to live(TTL) inforced.
  • Model distillation:
  • Prompt engineering: It’s important to choose right model and use right prompt according to the model.
  • Small language models: Recent small language models(SLM) perform really well compared to just a few months ago. However, the what SLMs can do and can’t do is clear.
  • Using GenAI to understand legacy codebase:

Assess

  • AI-friendly code design: A good code design for human also works well for AI code agents.
  • AI-powered UI testing: e.g. QA.tech, LambdaTests’ KaneAI, Browser Use
  • Structured output from LLMs: when LLMs response in JSON, it’s easier to use the result in automated system.

Hold

  • AI-accelerated shadow IT: I’m already seeing this happening, but it’s fair to say duct-tape solution can last long and can cause long-term harm than good.
  • Complacency with AI-generated code: As the speed of code generated by AI coding assistant is faster than developers can review, it’s easy for developers overlook the generated code. I personally did it for my personal work, and felt the pain that brings.
  • Local coding assistants: local LLMs’ performance and tooling capability is very limited compared to commercial AI models like ChatGPT, etc As of now, it’s only useful for PoC task or very simple tasks.
  • Replacing pair programming with AI: pair programming is about improving a whole team. AI can replace it.
  • Reverse ETL: Be extremely cautious to flow data from a central data platform to transaction processing system.
  • SAFe (Scaled Agile Framework): In short, be flexible and focus on value creation.

Platforms

Adopt

Trial

  • ABsmartly
  • Dapr: Distributed application runtime. Not useful for monolith simple app.
  • Railway: A full-stack alternative to Vercel.
  • Unblocked:  brings all the context about your codebase together, so your team gets expert-level answers, no matter where they’re working.
  • Weights & Biases: The AI developer platform to build AI agents, applications,
    and models with confidence

Assess

  • Arize Phoenix: Phoenix is an open-source AI observability platform designed for experimentation, evaluation, and troubleshooting.
  • Deepseek R1: DeepSeek-R1-Distill-Qwen-32B outperforming OpenAI-o1-mini on various benchmarks
  • Deno: Alternative to Node.js. Deno 2 made it compatible with Node.js and npm.
  • Graphiti: Graphiti is a framework for building and querying temporally-aware knowledge graphs, specifically tailored for AI agents operating in dynamic environments. This is worth considering when creating GraphRAG.
  • Helocone: a managed LLMOps platform
  • Humanloop: LLM evals platform for teams to ship AI products that succeed
  • Model Context Protocol (MCP): An open protocol that enables seamless integration between LLM applications and external data sources and tools. Just curious why MCP is not Adopt in Tech Radar.
  • Open Web UI: Nice tool, easy to set up. But when commercial AI service like ChatGPT is easy to use and enough capable and Claude, Gemini, etc. I’m not sure where Open Web UI can be useful.
  • pg_mooncake: a PostgreSQL extension that adds columnar storage and vectorized execution.
  • Reasoning models: Good for complex task, but general-purpose LLM is still better off for many tasks.
  • Restate: Restate is a lightweight runtime to turn AI agents, workflows, and backend services into durable processes. Focus on your logic, not failure mechanics.
  • Supabase: Great for MVP.
  • Synthesized: test data generation from production data.
  • Tonic.ai: test data generation.
  • turbopuffer: vector and full-text search engine
  • VectorChord: PostgreSQL extension for vector similarity search

Tools

Adopt

  • Renovate: dependency management tool which can automatically update dependencies.
  • Vite: front-end build tool

Trial

  • Claude Sonnet: Best LLM for coding.
  • Cline: VS Code extension work as AI Agent coding tool.
  • Cursor: VS Code forked AI first IDE
  • D2: open source diagram-as-code tool.
  • JSON Crack: VS Code extension that renders interactive graph from textual data.
  • MailSlurp: test email and SMS through API to MailSlurp.
  • Metabase: open-source BI tool.
  • NeMo Guardrails: an open-source toolkit for easily adding programmable guardrails to LLM-based conversational applications.
  • Nyx: a versatile semantic release tool
  • OpenRewrite: an open-source automated refactoring ecosystem for source code, enabling developers to effectively eliminate technical debt within their repositories.
  • Software engineering agents: agentic coding become more mature now. It should go Adopt in my option.
  • Tuple: remote pair programming tool.
  • Turborepo: a high-performance build system for JavaScript and TypeScript codebases.

Assess

  • AnythingLLM: The all-in-one AI app you were looking for. Chat with your docs, use AI Agents, hyper-configurable, multi-user, & no frustrating setup required. More RAG focused and multi-user capable compared to Open WebUI.
  • Jujutsu: a powerful version control system for software projects.
  • Mergiraf: Mergiraf can solve a wide range of Git merge conflicts. Configure Git to use Mergiraf instead of its default merge heuristics. This will enhance git merge, revert, rebase, cherry-pick and more. You can also keep Git’s original behaviour and manually invoke Mergiraf after encountering conflicts.
  • OpenRouter: a unified API for accessing multiple large language models. One particularly useful feature is its ability to bypass API rate limits.
  • Redactive: With Redactive, your organization is protected against AI-enabled data leaks.
  • V0: an AI tool for generating front-end code from a screenshot, Figma design or simple prompt.
  • Windsurf: I personally prefer Claude Code.

Languages & Frameworks

Adopt

  • OpenTelemetry: OpenTelemetry is a collection of APIs, SDKs, and tools. Use it to instrument, generate, collect, and export telemetry data (metrics, logs, and traces) to help you analyze your software’s performance and behavior. This is becoming a de-facto standard in this area.

Trial

  • Effect: a powerful TypeScript library for building complex synchronous and asynchronous programs. With Effect, code become a little like a functional programming language.
  • Hasura GraphQL engine: a universal data access layer that simplifies building, running and governing high-quality APIs on different data sources.
  • LangGraph: an orchestration framework designed to build stateful multi-agent applications using LLMs
  • MarkitDown: converts various formats (PDF, HTML, PowerPoint, Word) into Markdown
  • Prisma ORM: ORM for typescript in Node.js environment.

Assess

  • .Net Aspire: Aspire provides tools, templates, and packages for building observable, production-ready distributed apps.
  • Browser Use: an open-source python library that enables LLM-based AI agents to use web browsers and access web applications. It leverages Playwright.
  • CrewAI: a platform designed to help you build and manage AI agents. I’m curious that can be replaced by Claude Code sub agents.
  • FastGraphRAG: an open-source implementation of GraphRAG.
  • Presidio: It provides fast identification and anonymization modules for private entities in text such as credit card numbers, names, locations, social security numbers, bitcoin wallets, US phone numbers, financial data and more. It’s open-source by Microsoft.

Hold

  • Node overload: Don’t stick to Node.js which is great for IO-heavy workloads, but it’s struggling with compute-heavy, data-heavy workloads.

Leave a comment

Your email address will not be published. Required fields are marked *