Apurv Kochar

Designed a reusable protein and peptide screening pipelines that process candidate pools from millions of sequences to 10–50 prioritized designs through binding/selectivity prediction, structure prediction, docking, interface analysis, energy scoring, and Final Ranking. Built a end to end VHH (nanobody) and scFv affinity-maturation pipeline combining evolutionary sequence generation, inverse folding, developability filters, predicted-affinity ranking, structure modeling, and consensus energy scoring; validated candidate ranking using literature benchmarks and MD, including a published SARS-CoV-2 nanobody system. Benchmarked and extended multiple de novo antibody/binder design workflows across targets of varying difficulty, comparing structural confidence, target engagement, design success, runtime, and compute cost; progressed selected designs through refolding checks, Rosetta/FoldX/EvoEF2 scoring, MD, and downstream affinity maturation.

AI Research Engineer at Nucleo-Vir Therapeutics (2025-06 – Present)

• Designed a reusable protein and peptide screening pipelines that process candidate pools from millions of sequences to 10–50 prioritized designs through binding/selectivity prediction, structure prediction, docking, interface analysis, energy scoring, and Final Ranking.
• Built a end to end VHH (nanobody) and scFv affinity-maturation pipeline combining evolutionary sequence generation, inverse folding, developability filters, predicted-affinity ranking, structure modeling, and consensus energy scoring; validated candidate ranking using literature benchmarks and MD, including a published SARS-CoV-2 nanobody system.
• Benchmarked and extended multiple de novo antibody/binder design workflows across targets of varying difficulty, comparing structural confidence, target engagement, design success, runtime, and compute cost; progressed selected designs through refolding checks, Rosetta/FoldX/EvoEF2 scoring, MD, and downstream affinity maturation.
• Optimizing protein model inference using mixed precision, chunking, attention/memory tuning, and cache management and enabling prediction of 2,000+ residue complexes under constrained GPU memory, optimizing ESMFold single-chain inference achieving 2-3x faster inference.
• Established reusable AIRAWAT HPC workflows using Slurm and Enroot, automated previously manual ranking/reporting steps, and exposed computational pipelines through FastAPI endpoints for integration with internal agentic systems and use by bioinformatics and structural-biology teams.

LLM Safety / Generative AI Intern at DMI Finance (2024-07 – 2024-10)

• Built and extended an LLM red teaming and safety evaluation framework with Microsoft PyRIT and HarmBench, testing for jailbreaks, prompt injection, and other vulnerabilities using many automated attack strategies.
• Developed custom detectors for toxicity, jailbreaks, and personal data (PII), and built hallucination and fact-checking pipelines based on retrieval.
• Used automated evaluators and attack success metrics to study how results change with attack design, evaluator choice, and success criteria. Built Streamlit and Flask interfaces for testing and demos.

Generative AI Intern at AR Genie (2025-02 – 2025-04)

• Developed an agentic LLM workflow that generates, evaluates, and refines structured technical content over several iterative stages.
• Built a full web application that connects LLM orchestration, backend APIs, and the user facing workflow for technical-content generation.

Bachelor of Engineering in Computer Engineering – Savitribai Phule Pune University (2021 – 2025)