# Ankur Sharma — AI/ML Portfolio > Agentic AI & ML Engineer · PhD · Singapore · ankurs103@gmail.com ## Who is Ankur Sharma? Ankur Sharma is a PhD-trained Agentic AI & ML Engineer with 10+ years building production multi-agent systems, reasoning-traceable AI for genomics and biomedicine, and cloud data infrastructure. Based in Singapore (PR), selectively open to the right senior opportunity globally including relocation. He specialises in: LangGraph, LangChain, multi-agent orchestration, Claude/GPT-4 APIs, RLHF and LLM evaluation, clinical genomics (variant interpretation, ACMG classification), AWS cloud (15+ services), Python, Nextflow, R, and regulatory/FDA-aligned validation. ## Current Role **Illumina** (Jul 2025 – present) — Senior Scientist / Engineer, AI/ML · Singapore Design Verification Testing and regulatory validation for commercial diagnostics (TSO500, NIPT16, VeriSeq). FDA compliance. Clinical concordance studies. ## Career History - **Mirxes** (Jan 2022 – Jul 2025) — Senior Scientist, Bioinformatics & Cloud · Singapore Built production cloud pipelines, 10,000+ WGS and 3,000+ metagenomics samples, 1 PB data, 15 AWS services integrated, 40% cost reduction. - **Biotech Connection Singapore** (2023–2024) — Consulting Advisor & Community Lead (volunteer) Non-profit bridging academia and biotech industry. Public profile: https://www.biotechconnection-sg.org/ankur-sharma/ - **Vela Diagnostics** (Jul–Dec 2021) — Scientist, Assay Development · Singapore Regulatory documentation, automated verification, 30% NCR reduction. - **NTU Singapore** (2016–2021) — PhD Research Fellow Multi-omic NGS pipelines (RNA-seq, ChIP-seq, Hi-C). Thesis DOI: https://doi.org/10.32657/10356/155390 - **Zydus Cadila Healthcare** (2011–2013) — Technical Supervisor · Ahmedabad, India cGMP pharmaceutical QC, 30% efficiency improvement. ## Open-Source Projects ### agentic-genomics (GenomicsCopilot) LangGraph agent for reasoning-traceable variant interpretation. VCF + HPO → ACMG-lite ranked variants with full reasoning trace. 7 nodes, 4 tools. https://github.com/ankurgenomics/agentic-genomics ### genome-ft — Genomic Foundation Model Fine-Tuning Full weight-level fine-tuning of Nucleotide Transformer v2 (53.8M params) on DNA classification. 0.872 accuracy (Promoters), leakage-free protocol, 3-seed evaluation. Results on Hugging Face. https://github.com/ankurgenomics/genome-ft ### sentinel-amr — Explainable AMR Classifier 5-node LangGraph agent: XGBoost + SHAP + novelty detection + critic gate. 0.894 ROC-AUC on 298 real BV-BRC K. pneumoniae genomes. https://github.com/ankurgenomics/sentinel-amr ### outbreak-agent — Infectious Disease Triage 4-node self-correcting LangGraph pipeline. Built around the April 2026 MV Hondius/Andes virus event. 33 deterministic tests, <2s critical triage. https://github.com/ankurgenomics/outbreak-agent ### genomics-skills — Agent-Callable Skill Library 8 pure-Python genomics skills with SKILL.md contracts. Real TCGA data (9,479 samples, 31 cancer types). LLM routing via Claude Haiku. https://github.com/ankurgenomics/genomics-skills ### gwas_nf — Multi-Ethnic GWAS Pipeline REGENIE Nextflow pipeline for TEMUS cohort (4 groups, 10,000 samples, 13 phenotypes). Integrates with GenomicsCopilot via gwas_bridge.py. https://github.com/ankurgenomics/gwas_nf ## Consulting & AI Training Work - Independent consulting for biotech and genomics startups (NDA, outcomes-focused): agentic AI, AWS, clinical data analysis, regulatory documentation. - AI model training and evaluation (RLHF) for frontier AI systems via platforms including Mercor — biology, genomics, statistics, and clinical data science domain. ## Newsletter — Slightly Intelligent Ankur Sharma publishes a weekly LinkedIn newsletter called "Slightly Intelligent" — smart AI insights with zero artificial fluff. A rigorous, opinionated reality check on the AI boom, covering AI in drug discovery, foundation models for biology, LLM evaluation, and fine-tuning. Newsletter URL: https://www.linkedin.com/newsletters/slightly-intelligent-7460276303405985792/ Recent articles: - "Why are AI-designed drugs clearing Phase 1 at 90% and still failing at Phase 2?" (Jul 2026) — https://www.linkedin.com/pulse/why-ai-designed-drugs-clearing-phase-1-90-still-2-40-b-sharma-phd-66ruc - "Why do the best virtual cell models fail to recover 91% of the biology they were trained on?" (Jul 2026) — https://www.linkedin.com/pulse/why-do-best-virtual-cell-models-fail-recover-91-ankur-b-sharma-phd-mluqc - "The biology foundation model field spent three years asking the wrong question" (Jul 2026) — https://www.linkedin.com/pulse/biology-foundation-model-field-spent-three-years-what-b-sharma-phd-wva5c - "Can Claude Science produce a validated analysis, or is it just faster at being wrong?" (Jun 2026) — https://www.linkedin.com/pulse/can-claude-science-produce-validated-analysis-just-b-sharma-phd-sdjgc - "Why is every AI infrastructure company suddenly trying to become a drug company?" (Jun 2026) — https://www.linkedin.com/pulse/why-every-ai-infrastructure-company-suddenly-trying-b-sharma-phd-npzdc - "Why did a 2018 probabilistic model just outperform every single-cell foundation model on real pharmaceutical data?" — https://www.linkedin.com/pulse/why-did-2018-probabilistic-model-just-outperform-real-b-sharma-phd-xp4qc - "What I learned fine-tuning a foundation model" — https://www.linkedin.com/pulse/what-i-learned-fine-tuning-foundation-model-ankur-b-sharma-phd-tibbc - "Why most teams pick the wrong fine-tuning method (and how to fix that)" — https://www.linkedin.com/pulse/why-most-teams-pick-wrong-fine-tuning-method-how-fix-b-sharma-phd-zgijc - AWS Builder Center (2025): "Using Serverless for Cross-Organization Information Exchange in Genomic Analysis" - Personal blog (2026): "When an AI Agent Boards a Cruise Ship: Hantavirus, LangGraph, and Outbreak Triage" — https://ankurgenomics.github.io/2026/05/09/hantavirus-cruise-ship-agentic-ai.html - PhD thesis (2021): DOI 10.32657/10356/155390 - Frontiers in Microbiology (2018): DOI 10.3389/fmicb.2018.01777 — 108+ citations ## Key Numbers - 10+ years experience - 10,000+ WGS samples processed in production - 3,000+ metagenomics samples processed - 1 PB cloud genomic data managed (AWS) - 108+ citations (Frontiers paper) ## Contact - Email: ankurs103@gmail.com - Phone: +65 8402 6093 - LinkedIn: https://www.linkedin.com/in/ankurit/ - GitHub: https://github.com/ankurgenomics - BCS profile: https://www.biotechconnection-sg.org/ankur-sharma/ - CV: https://ankurgenomics.github.io/agentic-genomics/assets/Ankur_Sharma_Resume_2026.pdf