ADAMYA SINGH
> Agentic Engineer
ABOUT ME
> I build agentic systems, ML infrastructure, and polished full-stack products.
ML Engineer Intern at Aviro, building RL environments for enterprise AI agents.
Studied VLM negation failures in medical imaging as a Computer Vision researcher at Rutgers.
Software Engineering Intern at Moody's, re-architecting production data infrastructure.
SKILLSET
LANGUAGES
Core programming languages and scripting
AI / ML
Agentic systems, RL training, and applied ML
INFRA / BACKEND
APIs, retrieval systems, and deployment foundations
PRODUCT
Full-stack interfaces and production app delivery
PROJECTS
Built a production AI scheduling co-pilot for Rutgers students with a 22-tool Mastra agent spanning live SOC search, conflict detection, prerequisite resolution, and chat-driven UI mutations through Cedar-OS frontend tools. Embedded a Browserbase + Stagehand remote browser for credential-free Degree Navigator automation, persisting Zod-validated audit profiles to Supabase with JWT-derived ownership and RLS-hardened tables.
Built a MuJoCo + Gymnasium simulation environment for the SO-ARM-101 arm with multi-camera observations, normalized state/action interfaces, teleoperation, and shaped pick-and-place rewards. Implemented VLA policy inference with adapters for SmolVLA (450M) and Pi0 (3.3B), enabling model swapping via CLI flags. Trained policies with ReinFlow using PPO-style objectives.
Built a scalable AI-powered behavioral coach by fine-tuning Meta's LLaMA 3 8B on Atomic Habits, enabling high-quality, personalized habit formation guidance. Created a 2,300+ Q&A dataset and fine-tuned using LoRA and QLoRA, reducing memory requirements by 60%+ — making training viable on consumer hardware.
A modern, full-stack task management application built with React, TypeScript, and Spring Boot. This project demonstrates professional-grade web development practices, including type safety, optimistic UI updates, and responsive design. Features include task creation, real-time performance metrics, and error handling.
EXPERIENCE
MACHINE LEARNING ENGINEER INTERN
YC-BACKED- Built a Dockerized Enterprise Search Benchmark over PDFs, PowerPoints, and spreadsheets, backed by PostgreSQL + pgvector semantic search
- Designed an MCP-based RL environment with search/fetch/answer tools, step budgeting, citation validation, and redaction-after-view memory constraints, improving model performance by 26%
- Trained Qwen3-14B with GRPO using W&B serverless distillation and local on-policy RL with vLLM + LoRA hot-reloaded adapters
SOFTWARE ENGINEERING INTERN
DATA ENGINEERING- Re-architected a 24/7 ingestion platform by splitting 49 tables (~2.2 TB in production) from a PostgreSQL monolith and adding a dual-datasource Java/Spring layer across 5 repositories
- Built a restartable AWS RDS migration and verification pipeline, migrating ~255 GB / ~1.5B rows with 49/49 count parity, portable checksums, and entity-graph validation
- Designed replay-safe cross-database processing with idempotent reruns, preventing duplicate entities during failure recovery
SOFTWARE ENGINEER
FUSEN FELLOW- Engineered an enterprise LMS end-to-end: Spring Boot + PostgreSQL services with JWT auth and REST APIs, Next.js + TypeScript frontend with role-based dashboards
- Implemented AI-driven course generation pipelines reducing content creation time by ~80% while preserving quality across enterprise training programs
- Partnered with 15+ enterprise tech orgs to validate requirements and ship production-ready specs for large-scale deployments (10K+ learners)
UNDERGRADUATE RESEARCH ASSISTANT
COMPUTER VISION- Investigated vision-language model (VLM) failures in negation comprehension using 225K+ multimodal medical images from the CheXpert dataset
- Benchmarked 3+ GenAI VLMs (CLIP, BioMedCLIP) on negation-sensitive tasks using PyTorch; developed domain-specific evaluation tools
EDUCATION
RUTGERS UNIVERSITY
B.S. in Computer Science & Data Science
COURSEWORK
LEADERSHIP & CLUBS
USACS — Tech Team Lead & Executive Board Member
STAR — Space Technology Association at Rutgers