Research / systems / practice
I’m an IIT Delhi computer science graduate and Co-Founder & CTO of Lymph Labs. My work brings together biomedical NLP, retrieval-augmented generation, multi-agent systems, and the evaluation of model reliability.
My master’s research with Prof. Mausam studied medical model adaptation and diagnostic agents. Earlier, at Zuse Institute Berlin, I worked on asynchronous convex optimization. At Abacus.ai, I built agent capabilities and LLM infrastructure.
Explore the researchFine-tuning biomedical models, grounding answers in PubMed, and testing whether specialized agents improve diagnosis. A study of both performance gains and clinical failure modes.
Thesis experiments · MEDITRON-7B fine-tuning and revised diagnostic agents, respectively.
Methods, ablations & findings →An open-source implementation of a nonsmooth convex optimization algorithm, evaluated on sparse classifier training and stereoscopic image recovery.
Up to 24% lower runtime in the reported simulated-latency classifier experiment; studied convergence proxies, worker overhead, and hyperparameter schedules.
Implementation & experimental results →Implemented GPT-2 graph-to-text modeling from scratch and a GCN embedding experiment on WebNLG. The combined project reports BLEU 23.00 overall, with a substantial seen/unseen gap.
Compared classical classifiers and LLM prompting for Wikipedia controversy detection: 87% for subtopic-based prompts versus 93.1% for the strongest classical baseline on 29 test examples.
Research and applied work across healthcare AI, agents, optimization, and parallel computing. Open an entry for the full account.
Jul 2026 – Present
Co-Founder & CTO
Co-founded a medical evidence-search platform used by 2,000+ doctors, including clinicians from AIIMS and JIPMER. Led work across hybrid retrieval, citation-grounded LLM workflows, evaluation, durable operations, and web/mobile applications.
Jul 2025 – Jul 2026
ML Platform Engineer
Built the infrastructure framework for users to create and use custom skills with their own DeepAgent instances, alongside autonomous software engineering and multi-provider LLM infrastructure.
May – Jun 2025
Founding Engineer Intern
Built document and web ingestion workflows for recruiting data, combining file processing, scraping, search, and LLM-based extraction.
Jan 2024 – May 2025
Gen AI / NLP Research Intern · Master’s thesis
Investigated domain adaptation, retrieval, and multi-agent diagnosis for personalized treatment pathways under Prof. Mausam. The thesis was submitted in June 2025.
Jul 2024 – Jun 2025
Teaching Assistant · Parallel Programming
Mentored 150+ students in parallel programming and built the infrastructure to evaluate parallel computing assignments.
May – Jul 2024
Quant Intern · High-Frequency Trading
Studied equity derivatives and high-frequency arbitrage using volume, trades, and order-book dynamics.
May – Jul 2023
Software Developer / ML Intern
Worked on Ripik Optimus, an AI-driven production planner for SunPharma’s pharmaceutical manufacturing workflows.
Jun – Aug 2022
Research Intern
Co-authored an Optimization Online preprint and built asynchronous projective splitting in Julia for nonsmooth convex optimization, sparse classifier training, and image recovery.
A broader record of work in language modeling, machine learning, and systems.
NLP · Language model fine-tuning
Implemented GPT-2 graph-to-text modeling from scratch and a GCN embedding experiment on WebNLG. The combined project reports BLEU 23.00 overall, with a substantial seen/unseen gap.
NLP · Sequence labeling
Combined BERT representations with a conditional random field to identify hateful spans within sentences.
NLP · Oct – Nov 2023 · Prof. Abhijnan Chakraborty
Compared classical classifiers and LLM prompting for Wikipedia controversy detection: 87% for subtopic-based prompts versus 93.1% for the strongest classical baseline on 29 test examples.
Machine learning · Sep – Oct 2022 · Prof. Parag Singla
Implemented Naive Bayes for movie-review sentiment and a neural network for ten-class Fashion-MNIST classification.
Machine learning · GPU computing
Built a C++/CUDA image-processing library and implemented LeNet-5 for GPU-accelerated MNIST recognition.
Systems · Parallel algorithms
Generated and solved 64 × 64 perfect mazes using MPI and graph algorithms including BFS, DFS, Kruskal’s, and Dijkstra’s.
Systems · Concurrent editing
Built a collaborative Jupyter notebook-style editor using conflict-free replicated data types.
Systems · Feb – Apr 2023
Implemented scheduling, memory management, device drivers, and custom process-monitoring system calls.
B.Tech + M.Tech in Computer Science · 2020–2025
B.Tech GPA: 9.00 / 10
My interests span how models learn domain knowledge, how retrieval and agents affect reasoning, and how evaluation reveals weaknesses that aggregate scores can hide. A background in optimization and parallel computing informs how I approach these problems.
Machine Learning (COL774), Advanced AI (COL864), NLP (ELL884), Advances in LLMs (AIL821), Artificial Intelligence (COL333), and Data Structures (COL106).
Elsewhere: YouTube ↗
Let’s connect
I’m interested in conversations about AI research, medical NLP, retrieval, agent systems, and model reliability.
aryan.iitd1@gmail.com