ML/AI Systems Engineer & Technical Founder
New York, NY · ashesh8500@gmail.com · github.com/ashesh8500
Graduate student in computer engineering at NYU Tandon. Technical cofounder of AuditLayer. Work spans production LLM systems, spatial reasoning under pose noise, and empirical optimization research.
Competitive-intelligence software for wellness creators. Owns technical development: Next.js, Supabase/Postgres, Python workers, authenticated intake, queued agent runs, event logs, private report delivery, billing state, and founder review controls. Separated deterministic orchestration from agentic research with bounded execution, source provenance, and reproducible delivery.
Production LLM/SLM, semantic retrieval, and multimodal data workflows. Python, PyTorch, vector databases, Azure/AWS. RAG and semantic-memory systems in serverless product contexts.
LLM-powered document-intelligence and chat pipeline using retrieval, vector databases, Azure Blob Storage, and serverless infrastructure.
Longitudinal biomedical and neuroimaging analysis with Python, NumPy/Pandas, Statsmodels, GEE. Honors thesis: adolescent nicotine use and impulsivity, ABCD Study (N=9,562).
Leakage-controlled object-association experiment using runtime-available visual, geometric, and graph features. Evaluated 2,400 train / 186 validation / 771 test candidates under clean and pose-noisy conditions.
Class balance, strict-threshold coverage, pseudo-label precision, ablations, and failure modes reported separately. Code is private. project note
Paper-trading research system with auditable decision journal. Records observations, allocation, risk gating, and simulated execution. Profitability not established. project note
Screening, weighting, and turnover across three portfolio universes. Includes ablations and negative results. Top-100 best: Sharpe 1.47 (21d momentum + equal weight). Top-500 best: Sharpe 1.76 (63d vol-adj momentum + 21d rank-and-hold).
ML & Neural Computation specialization. GPA 3.75/4.00. Provost Honors. Honors thesis.
Python, PyTorch, scikit-learn, NumPy/SciPy, Pandas, Statsmodels, LLMs, RAG, evaluation, reinforcement learning, statistical modeling
Rust, C/C++, TypeScript/JavaScript, SQL, Linux, Docker, Git, gRPC, Supabase/Postgres, Azure, AWS, Cloudflare
GEE modeling, longitudinal analysis, neuroimaging, ABCD cohort, walk-forward evaluation, ablation design, reproducible CLI workflows