AI Strategy & MLOps
Moving beyond notebooks to production-ready AI. My focus is on the operationalization of LLMs and the automation of the ML lifecycle.
Frameworks & Orchestration
| Tool | Status | Why |
|---|---|---|
| LangChain | ADOPT | The foundational framework for LLM application development. Ubiquitous but heavy. |
| LangGraph | TRIAL | Stateful orchestration for building complex, cyclic agentic workflows. |
| Pydantic AI | TRIAL | Type-safe, model-agnostic agent framework with excellent developer experience. |
| CrewAI | TRIAL | High-level framework for orchestrating role-based autonomous agents. |
| DSPy | TRIAL | A new paradigm: Programming with prompts. Optimizes prompts algorithmically. |
| LangFlow | ASSESS | Visual prototyping tool for LangChain components. Good for quick demos. |
| LangServe | TRIAL | Deploys LangChain runnables and chains as REST APIs automatically. |
| Vertex AI Pipelines | ADOPT | Enterprise standard for managed ML workflows on GCP. |
| MLflow | ADOPT | The most mature platform for experiment tracking and model registry. |
| DVC | ADOPT | Data Version Control. Essential for versioning datasets and models alongside code (GitOps for Data). |
| OpenClaw | HOLD | SECURITY RISK: Autonomous agent with excessive permission scopes. Prone to Prompt Injection and RCE. Avoid in production. |
Inference & Serving
| Tool | Status | Context |
|---|---|---|
| vLLM | ADOPT | High-throughput and memory-efficient LLM serving engine. The production standard. |
| LiteLLM | ADOPT | Universal I/O for LLMs. Calls 100+ LLMs using the OpenAI format. |
| Ollama | ADOPT | Critical for local benchmarking, privacy-compliant testing, and edge LLM use. |
| OpenRouter | TRIAL | Unified interface for accessing top-tier models (Claude 3, GPT-4, Llama 3) with unified billing. |
RAG & Vector Search
| Component | Status | Context |
|---|---|---|
| LlamaIndex | ADOPT | My preferred framework for complex data ingestion and RAG optimization. |
| Hugging Face Transformers | ADOPT | The fundamental library for accessing state-of-the-art pre-trained models. |
| Pinecone | ADOPT | Managed vector database. Reliable, scalable, and easy to integrate. |
| ChromaDB | TRIAL | Open-source embedding database. Excellent for local development and testing. |
| Neo4j GraphRAG | TRIAL | Combining Knowledge Graphs with Vector Search for grounded, context-aware retrieval. |
LLM Evaluation & Observability
| Tool | Status | Why |
|---|---|---|
| LangSmith | ADOPT | Best-in-class observability for LLM apps. Essential for debugging chains and agents. |
| Ragas | ADOPT | Metrics-driven development. Essential for quantifying RAG performance (Faithfulness, Relevance). |
Memory & Protocols
| Component | Status | Why |
|---|---|---|
| Mem0 | TRIAL | The memory layer for personalized AI. Enables long-term user context retention. |
| OWASP Top 10 for LLM | ADOPT | Essential security standard for preventing injection and data leakage in GenAI. |
AI Coding Assistants
| Tool | Status | Context |
|---|---|---|
| GitHub Copilot CLI | ADOPT | Terminal-native assistant. Essential for explaining shell commands and git workflows quickly. |
| Codex (OpenAI) | ADOPT | Agentic coding partner for autonomous tasks, refactors, and code reviews. |
| Claude Code | TRIAL | Agentic CLI for complex refactoring tasks and autonomous codebase navigation. |
| TestSprite | TRIAL | AI-powered autonomous testing agent that generates and executes end-to-end tests. |
| Gemini CLI | ASSESS | Open-source alternative for GCP-integrated workflows. |
Classical ML & Data Science
| Tool | Status | Context |
|---|---|---|
| XGBoost / LightGBM | ADOPT | Gradient boosting frameworks for high-performance tabular data modeling. |
| TensorFlow / Keras | HOLD | Maintaining for legacy systems, but shifting focus to PyTorch and JAX. |
| Optuna | ADOPT | Automatic hyperparameter optimization framework, highly efficient and flexible. |
| Streamlit | ADOPT | Fastest way to build and share data apps. My default for internal tool prototyping. |
| Matplotlib / Seaborn / Plotly | ADOPT | Core visualization libraries for static and interactive charts. |
| SciPy | ADOPT | Fundamental library for scientific and technical computing. |
| SpaCy / NLTK | ADOPT | Industrial-strength NLP and classic text processing tools. |
| OpenCV | ADOPT | The comprehensive library for computer vision and image processing. |
| Imbalanced-learn | ADOPT | Essential for handling datasets with significant class imbalance. |