ML/AI Systems Engineer
New York, NY · ashesh8500@gmail.com · github.com/ashesh8500
The through-line: making machine uncertainty legible and reducible — scoped persistent memory for agents, traceable execution, and honest evaluation. Applied in production systems, spatial reasoning research, and an independent benchmark program.
venture · technical cofounder · 2026 – present
Production brand-intelligence workspace: Next.js, Supabase/Postgres, Python workers. Deterministic orchestration kept separate from agentic research — bounded execution, source provenance, reproducible delivery.
research infrastructure · solo · active
Cognitive investing layer: agentic research with explicit correctness guardrails, an auditable decision journal, and live + paper evaluation. Every position carries provenance; every verdict is worked or failed, with ablations. Profitability not established.
Independent public benchmark: can an agent retain and correctly use scoped persistent context across a long, distractor-filled task? Ships with a task definition, honest baselines, an eval set, ablations, and failure modes. project note
Screening, weighting, and turnover across three portfolio universes, with ablations and negative results. Best Top-100: Sharpe 1.47 (21d momentum + equal weight). Best Top-500: Sharpe 1.76 (63d vol-adj momentum + 21d rank-and-hold). project note · full study
NYU Wireless · 2026
Leakage-controlled experiment on runtime-available visual, geometric, and graph features: 2,400 train / 186 validation / 771 test candidates, evaluated under clean and pose-noisy conditions.
project note · code private · class balance, ablations, and failure modes in the note
Consulting AI Engineer, SageX Global. Production LLM/SLM systems, semantic retrieval, multimodal data workflows. Python, PyTorch, vector databases, Azure/AWS.
Machine Learning Intern, UniQreate. LLM document intelligence and chat pipeline: retrieval, vector databases, serverless deployment.
Undergraduate Research Assistant, UC San Diego. Longitudinal biomedical and neuroimaging analysis; honors thesis on adolescent nicotine use and impulsivity, ABCD Study (N=9,562).
MS Computer Engineering, NYU Tandon School of Engineering (expected).
BS with Honors, Cognitive Science, UC San Diego. ML & Neural Computation specialization. GPA 3.75/4.00. Provost Honors.