B.Tech Bioinformatics student (Class of 2027) who builds data and machine-learning pipelines end to end: acquiring public datasets, cleaning and representing them, training and validating models, and packaging the result so it runs with one command. Built a retrieval-augmented generation service with a provider-agnostic inference layer, and a reproducible multi-omics analysis pipeline now prepared for journal submission. Looking for a DevOps / MLOps role where reliable data preparation, deployment and model monitoring matter.
Languages: Python (primary), R, C, JavaScript/TypeScript, SQL, shell
ML / NLP: scikit-learn, sentence-transformers, LangChain, vector search, embeddings, RAG, evaluation (CV, C-index, decision curves), PyTorch basics
Data & pipelines: reproducible one-command pipelines, public-data ingestion (TCGA, GEO, GTEx, ChEMBL), batch-effect diagnosis, Git, Linux
Serving: Streamlit, OpenAI-compatible LLM endpoints (Ollama, vLLM), streaming responses
Fundamentals: OOP, data structures & algorithms, biostatistics
Python | LangChain | sentence-transformers (all-MiniLM-L6-v2) | vector store | Ollama / vLLM | Streamlit
- Built the full pipeline: PDF/text ingestion, chunking, embedding, semantic retrieval, and streamed grounded answers with inline [n] citations and relevance-scored sources.
- Designed a provider-agnostic LLM layer so the same code serves from a local Ollama model or a hosted vLLM endpoint, switching by configuration only.
R 4.3 | TCGA-STAD, GTEx, 5 GEO cohorts, scRNA-seq, 16S microbiome, IEU OpenGWAS | LASSO-Cox, WGCNA, GSEA, Mendelian randomisation
- Acquired, harmonised and versioned eight public datasets into a reproducible one-command R pipeline; trained a 25-gene LASSO-Cox prognostic model with nested cross-validation (C-index 0.61) and validated it on independent external cohorts.
- Diagnosed an apparent microbiome “dysbiosis” signal as a sequencing-batch artefact through drift analysis across cohorts; reported an honest null from two-sample Mendelian randomisation. Manuscript prepared for journal submission.
Cracking KRAS G12D — AI-Guided De Novo Inhibitor DesignBoltzmann AI Hackathon 2026, individual entry
Python | DiffDock | RDKit | ChEMBL / SureChEMBL | PDB 7RPZ
- Generated and filtered 46 de novo molecules against the KRAS G12D switch II pocket with an approved inhibitor (MRTX1133) as positive control through every stage; 8 of 15 candidates beat the reference’s in-pocket pose confidence, reproduced across two independent runs.
- Automated the screen as a scripted pipeline: docking, drug-likeness (Lipinski / QED), and novelty checks against ChEMBL (Tanimoto 0.43–0.67).
Smart India Hackathon, cleared internal college selection | JavaScript | Node.js | LLM chatbot with voice
- Built the frontend: mood-adaptive chatbot interface, anonymous peer connect and one-click personalised actions.
OpenAI Hackathon, “AI in Daily Life” track | TypeScript
- Team lead. Owned the concept, split the work across the team and drove the build to a working hackathon submission.
B.Tech, Bioinformatics — SASTRA Deemed to be University, ThanjavurExpected 2027
Data Structures & Algorithms, Machine Learning, Deep Learning, Big Data, Biostatistics, Genomics & Proteomics
Class XII & X — MES Central School, Tirur, Kerala78.4% (2021) | 89.4% (2019)