llm-evaluations

A hyperlocal job board purpose-built for India's Tier-2/3 cities — employers post geotagged jobs, candidates upload resumes that an LLM auto-parses into skill vectors, Elasticsearch matches candidates to nearby jobs within a configurable radius, and job alerts arrive on WhatsApp in the candidate's regional language.

Evaluation suite for a banking RAG assistant — faithfulness, hallucination & relevance scoring with CI regression gates (DeepEval · RAGAS · pytest)

benchmarking jailbreak-dataset fine-tuning-tools llm-evaluation official high-priority benchmarking task evaluation checkpoint shared memory sandbox API endpoints structured data schemas collaborative environments

Production-ready enterprise RAG and structured data extraction pipeline using LangGraph, Instructor, and Pydantic with automated CI evaluation metrics.

Replication-first study of sociolinguistic effects on LLM epistemic judgments, with behavioral validity gates before mechanistic interpretation.

Defense-in-depth safety study on Mistral-7B: how much of an LLM's safety lives in the weights vs. external guardrails, and how fragile it is under fine-tuning. Defensive research.