Looking for a job

Ninad Naik

@ninadnaik

AI engineer at Recollia building reproducible LLM evaluation workflows to audit reasoning reliability and hallucinations.

India

Message

At Recollia, I design evaluation workflows to audit LLM reasoning reliability, instruction-following robustness, and hallucination failure modes. I’m also building reproducible benchmarks to stress-test model behavior across complex multi-turn tasks.

I developed Mirage, a modular platform for measuring hallucination risk across local and API-hosted models, and NirikshaID Bench, a 600-document benchmark evaluating VLM extraction, tamper detection, robustness, and calibration. My other work includes a PPO-based Mars landing controller with deterministic stress testing and a PyTorch CNN-LSTM manuscript on forecasting crop stress from satellite and climate data.

Experience

Work history, roles, and key accomplishments

Current

AI Engineer Intern

Current

Recollia

Oct 2026 - Present (0 months)

Designing and executing evaluation workflows to audit LLM outputs for reasoning reliability, instruction-following robustness, and hallucination failure modes. Building reproducible benchmarking pipelines to stress-test model behavior and measure output consistency across complex multi-turn tasks.

LLM Evaluation Hallucination Auditing Benchmarking Python

AI Engineering Intern

ElevateSoul.ai

Aug 2026 - Sep 2026 (1 month)

Built a LangGraph-powered conversational AI evaluation pipeline, designing multi-stage agent workflows for structured candidate scoring and tool selection. Implemented a RAG evaluation pipeline over a custom knowledge base, grounding model outputs in retrieved evidence and systematically auditing response consistency.

LangGraph RAG LLM Evaluation Python

AI Engineering Intern

Fanisin

May 2026 - Jul 2026 (2 months)

Engineered conversational memory agents with persistent context and systematically evaluated generative outputs for hallucinations, context drift, and failure edge cases.

Conversational AI LLM Evaluation Python

Data Science Intern

ZeTheta Algorithms Pvt. Ltd.

Feb 2026 - Mar 2026 (1 month)

Built autoencoder-based anomaly-detection pipelines across 50k+ high-dimensional records, conducting rigorous failure analysis against baselines to reduce false positives by 30%.

Autoencoder Anomaly Detection Python scikit learn