Ankit Singh Chauhan

Ankit Singh Chauhan

AI Researcher — Large Language Models (agents, RAG, PEFT/QLoRA), Causal Inference and Probes, Knowledge Graphs

Projects

Selected applied-ML and research engineering work — LLM fine-tuning, RAG pipelines, evaluation frameworks, and production deployments.

LLM-Driven Community Asset Knowledge Graph (Mistral-7B + Bedrock)

· AI Researcher · Indiana University Indianapolis

Two-pass LLM extraction pipeline on Amazon Bedrock Agents — Haiku classifier routes documents between Sonnet and a LoRA-fine-tuned Mistral-7B — lifting structured extraction from 0.32 → 0.78 Triple-F1 over 8.3k records, with a React/Cytoscape.js/FastAPI HITL review UI validated at 0.82 Cohen's κ.

  • LLM
  • LoRA
  • Mistral-7B
  • RAG
  • Amazon Bedrock
  • Lambda
  • DynamoDB
  • FastAPI
  • React
  • Cytoscape.js
  • Neo4j

CultureEval: Quantifying Cultural Alignment in LLMs

· Academic Project · Indiana University Indianapolis

PCA-based evaluation framework over ~97k survey respondents and 96 sociocultural indicators, surfacing systematic underestimation of religious-traditional values in Llama-2, Gemma-3, and Phi-4 (Cohen's d 0.89–1.17).

  • LLM evaluation
  • PCA
  • Llama-2
  • Gemma-3
  • Phi-4
  • Cohen's d

Pedagogical Instruction-Tuning for Llama-3.2 — NSF CATpc

· Research Assistant · Indiana University Indianapolis (NSF-funded)

Built a 7.2k high-quality instruction-tuning dataset (Cohen's κ 0.88) and fine-tuned Llama-3.2 with QLoRA + DPO for the NSF-funded CATpc project — +14% pedagogical alignment, −8% hallucinations vs. baseline.

  • LLM fine-tuning
  • QLoRA
  • DPO
  • Llama-3.2
  • dataset construction
  • RLHF
Code NSF Award

Chart-to-Text: CNN-RNN vs Vision-Language Models

· Course Project · Indiana University Indianapolis

Swapped ResNet-50 for EfficientNet-B2 and added dual-LSTM decoding with coverage attention on 27k+ Statista chart images, lifting BLEU-4 from 0.18 to ~0.50.

  • PyTorch
  • EfficientNet
  • LSTM
  • coverage attention
  • BLEU-4
  • VLMs