Hybrid BM25 + dense retrieval with RRF fusion. Navigation-first RAG for long documents.

hybrid retrieval, R@3 0.956 over 500 SQuAD questions — benchmark script and results in-repo

Stop RAG from failing silently. Query rewriting, retrieval confidence scoring, fallback strategies, retry logic.

Multi-agent orchestration. ReAct, Swarm, Pipeline, Debate, WorkflowGraph patterns. Routing, planning, RAG, cost tracking.

Agentic RAG on Nepal government policy and legal documents. Hybrid retrieval with citations, offline-capable, Nepali + English.

Voice AI for healthcare. Call intake, structured extraction, missing-data recovery, appointment orchestration, post-call analytics.

Playwright-based web scraping. LLM extraction, hybrid selectors, session persistence, rate limiting, anti-detection.

Natural-language queries on CSV data using local LLMs. No API keys, no cloud.

Lightweight NLP utilities bridging traditional pipelines and transformer-ready workflows.

Agentic feature engineering. PyTorch + LangChain agents for multimodal feature extraction.

Deep learning toxicity detection. Per-category scores for obscene language, threats, insults, identity hate.

Extract social media profile links from websites. Useful for lead enrichment and web intelligence pipelines.

Scrape Trustpilot reviews programmatically. Structured output for sentiment analysis and competitive research.