3. Data, Science & AI

Generates query variants, retrieves results for each, and fuses them using RRF to improve search quality for ambiguous or under-specified queries.

Builds Retrieval-Augmented Generation systems with document ingestion, hybrid search, reranking, and prompt augmentation for grounded LLM responses.

Provides foundational guidance for document storage, search, and RAG system implementation using llmemory, including vector, semantic, and BM25 techniques.

Enables optimized search retrieval by fusing vector and keyword results via Reciprocal Rank Fusion and alpha tuning for hybrid search systems.

Compresses and simplifies AI prompts to preserve core meaning while minimizing context length for efficient language model processing.

Verifies claims in AI-generated content against external sources to detect and correct hallucinations, ensuring factual accuracy.

Generates a mobile-friendly single-file HTML GitHub Wrapped report with Bilibili-style narrative and embedded data visualization.

Analyzes YouTube comments for sentiment, topics, and trends. Integrates with video summaries for cross-analysis.

Converts PDF, DOCX, PPTX, images, audio, and other formats to Markdown for AI processing and text analysis.

Summarizes YouTube videos by transcribing audio with whisper.cpp and generating concise summaries using Claude CLI, providing a repeatable workflow for high-quality video content analysis.

Analyzes CSV data files by applying lookup guides and reference documentation for accurate interpretation and processing.

Generates AI-synthesized summaries of GitHub activity (issues, PRs, commits) for organizations and time periods, delivering natural language narratives with URL references.

Extracts HTML, links, text, and resources from web pages for data collection and analysis.

Constructs optimized document embedding pipelines for RAG systems, featuring text chunking, embedding generation, and indexing.

Synthesizes perspectives from multiple LLMs (e.g., ChatGPT, Gemini) to generate a cohesive implementation plan upon user request.

Generates clean, chunked JSONL datasets from documents for vector embeddings, including deduplication and quality checks.

Builds LLM applications with LangChain, including chains, agents, memory, tools, and RAG pipelines for conversational AI and data retrieval.

Builds end-to-end RAG pipelines for document-based AI assistants with customizable chunking, retrieval, and evaluation strategies.

Builds reusable evaluation frameworks for LLMs and AI models using golden datasets, scoring rubrics, and regression reports for benchmarking and quality assurance.

Creates reusable prompt templates with output contracts, style rules, and quality criteria for consistent LLM responses.

Designs multi-step AI agent workflows with tool orchestration, state management, retry logic, and cost controls for autonomous systems.

Optimizes SQL queries using EXPLAIN analysis, index recommendations, and performance benchmarking to enhance database efficiency.

Orchestrates multi-agent AI systems with task delegation, agent communication, shared memory, and workflow coordination for autonomous agent networks.

Compares historical and new prompts across test cases, providing diff summaries, stability metrics, breakage analysis, and fix suggestions for AI prompt regression testing.