Company Description

Moleculyst Ventures Private Limited is a deep-tech artificial intelligence research startup focused on building more reliable, autonomous, and scientifically capable AI systems.

Our current research focuses on hallucination detection in large language models, mechanistic understanding of model behavior, uncertainty and reliability estimation, and the development of automated AI research pipelines.

We are building systems that can assist with the full research cycle: reading technical literature, identifying open questions, generating hypotheses, designing experiments, executing computational studies, analyzing results, and retaining useful knowledge across iterations.

A central goal of our work is to understand when AI systems possess relevant internal knowledge but still produce incorrect or fabricated outputs, and to develop methods for detecting and reducing such failures.

Moleculyst is also developing infrastructure for self-improving research agents, including scientific memory systems, experiment orchestration, model evaluation, and tools for reproducible AI research.

Our broader objective is to create AI systems that can function as reliable scientific collaborators rather than straightforward language interfaces—systems capable of reasoning, experimentation, self-evaluation, and increasingly autonomous discovery.

Role Description — Natural Language Processing / AI Research Engineer

As a Natural Language Processing / AI Research Engineer at Moleculyst, you will work on research involving large language models, hallucination detection, model reliability, scientific reasoning, and automated AI research systems.





This role goes beyond conventional NLP tasks such as text classification or basic information extraction. You will study how language models represent and retrieve knowledge, why they hallucinate, and how internal model signals can be used to predict whether an output is reliable.

You will also contribute to the development of AI research agents capable of reading papers, generating hypotheses, running experiments, evaluating results, and improving their research strategies over time.

Key responsibilities include:

- Developing methods for detecting hallucinations and unreliable outputs from large language models.
- Analyzing transformer representations, hidden states, attention patterns, residual streams, and other internal model signals.
- Investigating whether models internally distinguish known information from uncertain or fabricated information.
- Designing controlled experiments and benchmark datasets for evaluating hallucination detection methods.
- Studying representation geometry, model confidence, uncertainty, and other signals related to factual reliability.
- Implementing and evaluating ideas from current research papers in NLP, language models, interpretability, and AI reasoning.
- Building automated research pipelines for literature analysis, hypothesis generation, experimentation, evaluation, and iterative improvement.




- Developing memory systems for research agents, including semantic, procedural, and experimental memory.
- Creating systems for storing research results, experimental observations, failed approaches, and reusable procedures.
- Building tools for experiment orchestration, automated evaluation, result comparison, and research reproducibility.
- Developing scientific literature processing and retrieval systems where required by the research pipeline.
- Working with local and remote large-language-model inference infrastructure.
- Maintaining high-quality experimental code, documentation, datasets, and research records.
- Collaborating with other researchers and engineers on open-ended AI research problems.

Qualifications

- Strong foundation in computer science, algorithms, data structures, and software engineering.
- Strong proficiency in Python.
- Solid understanding of machine learning and deep learning.
- Familiarity with transformer architectures and modern large language models.
- Practical experience with PyTorch and relevant scientific computing libraries.
- Understanding of embeddings, attention mechanisms, representation learning, and model evaluation.
- Ability to read research papers and independently implement experimental ideas.
- Strong experimental reasoning skills, including hypothesis formation, controlled evaluation, and interpretation of results.
- Familiarity with Git and reproducible research workflows.

Experience in any of the following is highly valuable:
- Hallucination detection, factuality evaluation, uncertainty estimation, or calibration.
- Mechanistic interpretability or transform

📌 Natural Language Processing Engineer (Delhi)
🏢 Moleculyst
📍 Delhi