headshot image

Anshumaan Singh

Machine Learning Systems

Interested in the full stack of machine learning — from the systems and infrastructure that run models at scale, to the research and training that makes those models better.

I want to work at the boundary where algorithms meet hardware constraints, where a smarter scheduler or a better architecture can make the difference. Whether that's optimizing inference pipelines, building distributed training systems, or pushing model quality through better fine-tuning — that's the work I'm drawn to.

Email Me Download Resume LinkedIn GitHub

About Me

~/about/personal-intro

PI Logo | Personal Introduction

I'm a Computer Science Honors student at Stony Brook University with a deep passion for Artificial Intelligence and Machine Learning.

My journey over the last three years has been a rapid cycle of learning, building, and adapting. I thrive in environments where I can dive headfirst into challenges and take ownership from start to finish.

At Bank of Montreal, as the sole engineer, I built a 39-tool Model Context Protocol server that cut a 3–4 hour manual AML investigation workflow to 1–2 minutes. At Stony Brook's LUNR AI Lab, I fine-tuned CodeLlama-7B, improving Coding RAG accuracy by 5.4% and achieving 73.5% faster benchmark runtimes.

What drives me isn't just technical skill—it's a relentless work ethic and genuine curiosity. When frustrated with generic phone notifications, I taught myself Kotlin and built NotiSentry, an AI-powered notification filter. This scrappy, problem-solving approach defines how I tackle challenges.

I'm looking for opportunities where I can contribute to a larger vision, learn from talented teams, and apply my "builder, owner, partner" mindset to real-world problems. Let's connect!

~/about/skills

Skills Logo | Skills

• Languages: Python, C, C#, Java, JavaScript, Kotlin (Android Jetpack Compose), OCaml, MATLAB, MIPS Assembly, HTML/CSS.

• Agentic AI & ML Systems: Model Context Protocol (MCP), RAG, LLM fine-tuning (LoRA/QLoRA), vLLM, PyTorch, TensorFlow, Transformers, LangChain, HuggingFace, Reinforcement Learning.

• Backend, APIs & Protocols: FastAPI, Flask, Node.js, React.js, Bootstrap, REST APIs, WebSockets, SSE, SQLAlchemy, Google Cloud, Firebase.

• Data, ML Tooling & Platform Engineering: Pandas, NumPy, OpenCV, PostgreSQL, MongoDB, MySQL, SQLite, DuckDB, Apache Parquet, ETL pipelines, AWS (EC2/S3), Docker, Linux/Unix, HPC Clusters (Slurm).

• Developer Tools & Practices: Git, GitHub, CI/CD, Playwright, RBAC/JWT auth, audit logging, Atlassian Suite.

~/about/education

SBU Logo | Education

• Degree: B.S. in Computer Science Honors

• University: Stony Brook University, New York, USA

• Expected Graduation: May 2027

• GPA: 3.9/4.0

• Honors: SUNY SOAR Fellow, URECA Fellow, Computer Science Honors, University Scholars, Dean's List (2023 – 2025)

• Awards: Academic Achievement Award, YouAreWelcomeHere Award, Global Excellence Award

• Relevant Coursework: Data Structures, Foundations of Computer Science, Systems Fundamentals I & II, Theory of Computation (Honors), Software Development, Research in Computer Science, Applied Calculus II & III, Applied Linear Algebra, Survey of Probability and Statistics

~/about/hobbies

Hobbies Logo | Hobbies

• Photography: I love taking photographs of landscape, nature and the nightsky. Check out some of the pictures I have taken on my Instagram!

• Sailing: I recently started sailing with the Stony Brook University's Sailing Club and I am a huge fan now. I absolutely love sailing!

• Sketching: I recently bought a new iPad Air and I have been learning how to sketch on it. So far... I'd say for someone who has never touched a sketch pen before, I am doing pretty decent.

• Video Games: When I am not taking photographs and editing them, not sailing or sketching then I am probably playing video games. It is one of my favourite indoor passtimes.

Work Experience

~/work/bmo
AI & ML Engineering Intern | June 2026 – August 2026

Bank of Montreal (BMO) | Berkeley Heights, NJ

• As sole engineer, built a 39-tool Python/FastAPI Model Context Protocol (MCP) server for an integrated AML case-investigation pipeline from a blank scaffold in ~6 weeks, implementing HTTP, SSE, and WebSocket transports.

• Designed the end-to-end AML pipeline so cron-scheduled backend jobs refresh the cache and agents call MCP tools on demand; reduced a 3–4 hour manual workflow to 1–2 minutes (~99% faster), with management projecting ~$2M in savings over two years.

• Engineered a three-pool concurrent job router with bounded queues, HTTP 429 back-pressure, and serialized cache writes; built the cron-scheduled cache-refresh service with atomic Parquet writes and retry/backoff.

• Built a C# adapter layer and automated ingestion pipelines to normalize heterogeneous upstream data for an analytics-platform modernization; added black-box regression tests with verified input/output fixtures to validate mappings against the legacy implementation.

~/research/reliable-systems-lab
Undergraduate Research Assistant | January 2026 – Present

Reliable Systems Lab | Stony Brook University, NY

Researching multi-agent cyber-physical systems — neural controllers and MPC algorithms for autonomous UAV-swarm coordination — advised by Prof. Scott Stoller and Prof. Scott Smolka.

• Engineering a lightweight encoder-only attention architecture for multi-agent systems, leveraging permutation equivariance to process jagged input arrays and eliminate data sorting overhead.

• Improving UAV Agent target seeking efficiency by 34.6% by architecting a Curriculum Weight Tuning API using CMA-ES through a grouped parameter strategy.

• Enabling 100% collision-free horizontal scaling on the SeaWulf HPC cluster by building a distributed isolation framework using dynamic sandboxing to eliminate MATLAB/MEX race conditions.

~/work/mailgator
Software Developer Intern | September 2025 - January 2026

Mailgator | Palo Alto, CA (Hybrid)

• Reduced validation latency by 99.9% (60m to <2s) by architecting a scalable CI/CD pipeline on AWS EC2 with automated asynchronous integrity checks for a FastAPI backend.

• Optimized system reliability by refactoring LLM-driven parsing logic for deterministic data extraction, ensuring 100% accuracy across complex sender-recipient edge cases.

• Developed Generative AI email-analysis systems with a five-person agile team, including prompt-QA and UX-test infrastructure for an ML backend built with FastAPI, Node.js, React, and PostgreSQL.

~/research/lunr-ai-lab
Undergraduate Research Assistant | February 2025 – March 2026

LUNR AI Lab | Stony Brook University, NY

• Improved Coding RAG accuracy by 5.4% and accelerated code RAG benchmark runtimes by 73.5% by fine-tuning CodeLlama-7B via LoRA and designing a parallelized vLLM routing system.

• Reduced LLM inference costs by up to 100% by architecting and integrating an SQLite3 caching layer directly into the model distillation pipeline.

• Fine-tuned CodeLlama-7B on a custom dataset of 460K+ samples; the resulting Coding RAG work improved MBPP evaluation accuracy by 5.4% and ODEX evaluation accuracy by 1.6%.

~/volunteer/humanity-unleashed
Volunteer AI Researcher & Data Engineer | November 2024 - February 2025

Humanity Unleashed (humun.org) | Remote

Volunteered at Humanity Unleashed, a self-funded volunteer-only research organization and contributed in the following ways:

• Built pipelines processing 700k rows of economic data, publishing 5 datasets to HuggingFace with 1,400+ downloads.

• Collaborated with 2 other peers to develop a policy generation and summarization pipeline utilizing LangChain to translate complex economic data into nuanced policy explanations.

~/academic/teaching-assistant
Teaching Assistant | January 2024 - December 2024

University Scholars Fellowship | Stony Brook University, NY

• Demonstrated leadership & mentorship skills as a leader for incoming University Scholars students at Stony Brook University.

• Acquired valuable teaching experience as a Teaching Assistant as part of the SCH 275 pre-fellowship program.

• Assisted the University Scholars Director as a Teaching Assistant by bringing my own life experiences and teaching style into the class.

• Designed creative presentations, graphics & videos to make weekly student workshops engaging and fun.

~/research/igem
Software Engineer (Generative AI & Full Stack) | February 2024 - October 2024

iGEM, Stony Brook University | Stony Brook, NY

• Achieved 90% retrieval accuracy on embedded research documents by building a RAG Q&A chatbot with Transformers and LangChain, improving the research-wiki experience.

• Led a three-person team to develop a Flask research wiki that attracted 15+ stakeholders, helping secure $50K+ in funding and expand outreach.

• Awarded a Gold Medal for the team's research efforts at the iGEM Paris 2024 Jamboree.

~/work/it-assistant
IT Assistant Intern | January 2024 - July 2024

Faculty Student Association at Stony Brook University | Stony Brook University, NY

• Developed and maintained databases using MySQL and Google Sheets, automated processes by building Google AppScript algorithms like Binary Sort and Search, improving automation and work efficiency.

• Inventoried, performed maintenance and resolved issues in over 400 systems across the university campus.

• Installed and configured computer hardware & software, including Windows and Linux operating systems, in 40+ systems.

~/work/omniscience
Full Stack Web Developer Intern | April 2023 - June 2023

Omniscience Corporation, Palo Alto, CA | Remote

• Created a life insurance flask WebApp with 3000+ lines of code using HTML, CSS, JavaScript, Jinja, MySQL and Bootstrap Library.

• Containerized the application using Docker and deployed it on Amazon Web Services to ensure a reliable & scalable environment.

• Improved accessibility using Bootstrap HTML Framework, prioritizing user experience and effective communication.

Featured Projects

Highlighting my most impactful work in AI, systems programming, and full-stack development

~/projects/llm/pi-mtplx
pi-mtplx (Local LLM Runtime Integration)

Published pi-mtplx, a Pi extension for local MTPLX inference that reached 1,600 downloads in its first two weeks, with zero-configuration discovery of locally installed models that derive context-window, vision, reasoning, and token-limit capabilities from each installed artifact.

Engineered a local model-server lifecycle with health checks and ownership-aware cleanup, persistent SSD session cache, thermal fan-mode controls, and live tokens-per-second monitoring in Pi's footer.

GitHub Project
~/projects/tools/pi-apply-job
pi-apply-job (Evidence-Backed Resume Automation)

Built a Pi extension that accepts individual or batched public job URLs, extracts job requirements, and creates isolated per-application artifacts — without ever submitting applications.

Designed an evidence-linked resume-plan and verification workflow where every selected resume claim cites a stable ID from a master resume, rendered deterministically to a validated one-page LaTeX PDF via Tectonic.

GitHub Project
~/projects/systems/fuzzer
Automated Systems Fuzzer

Engineered a high-performance C fuzzer utilizing Unix signals and syscalls (fork, waitpid) to stress-test executables via mutated input streams, achieving 100% process isolation and automated memory leak detection.

~/projects/tools/cmdflow
CMDFlow at HackPrinceton

Built a local-first, AI-powered command-tracking system (FastAPI, ReactJS, MongoDB) that streams shell activity (<1s latency), performs PII scrubbing, and semantically indexes commands for natural language search and automatic project-based organization.

GitHub Project Devpost
~/projects/ml/notisentry
NotiSentry: A Smart DND with LLMs

Optimizing system performance to <1% battery drain over 5 hours and achieving 1.3s worst-case latency for an intelligent notification filtering and summarization app to minimize user distractions and improve focus using Firebase Gemini API and Jetpack Compose.

~/projects/ml/replug-lsr
REPLUG LSR with vLLM

Refactored research implementation of REPLUG to enable LM-Supervised Retrieval (LSR) fine-tuning for code generation tasks by architecting a high-performance training pipeline with a local vLLM server.

GitHub Project (Coming Soon)

Additional Projects

~/projects/low-level/red-black-tree
Red-Black Tree in MIPS Assembly

Implementation of core functions for a Red-Black Tree data structure, written entirely in MIPS assembly language, focusing on low-level memory operations, register conventions, and complex balanced data structures.

GitHub Project
~/projects/systems/fs-emulator
C Based Linux Filesystem Emulator

In-memory emulation of a Linux-like filesystem written entirely in C, managing core filesystem structures like i-nodes and data blocks, handling memory allocation, and implementing a hierarchical file and directory system from the ground up.

GitHub Project
~/projects/networking/poker-server
Multi-Client Texas Hold'em Poker Server

Multi-threaded poker server built in Java supporting concurrent client connections, game state management, and real-time gameplay with robust error handling and network communication.

GitHub Project
~/projects/games/skyscrapers
Skyscrapers Puzzle Game

Implementation of the single-player puzzle game "Skyscrapers" in C, including both an interactive version and an automatic solver based on logical heuristics.

GitHub Project
~/projects/security/aflent-protocol
AFLENT Protocol and Custom Encryption

Low-level data manipulation in C, implementing a custom network protocol (AFLENT) and a block cipher for data encryption/decryption with byte and bit-level computations.

GitHub Project
~/projects/vision/sobel-operator
Sobel Operator Edge Detection

Image processing program implementing the Sobel operator algorithm to detect edges of objects in images.

GitHub Project
~/projects/nlp/text-similarity
Text Similarity Check

Java program for comparing text files and checking authorship similarity using cosine similarity algorithms.

GitHub Project
~/projects/algorithms/social-graph
Social Media Graph

Java simulation of a social media network using graph data structures to model followers and followings.

GitHub Project
~/projects/vision/box-blur
Box Blur Image Filter

Image processing program applying box blur algorithm to images, optimized for images under 800x800 pixels.

GitHub Project
~/projects/vision/bw-filter
Black & White Filter

Image filter applying black and white conversion by averaging RGB values of each pixel.

GitHub Project
~/projects/security/playfair
Playfair Encryption

Java implementation of the Playfair cipher for encrypting and decrypting text.

GitHub Project
~/projects/systems/c-tracer
C Program Tracer

Java tool for tracking C code blocks and variables initialized/updated within them.

GitHub Project
~/projects/java/hiring-sim
Hiring System Simulator

Java simulation of a complete hiring system manager for job applicants.

GitHub Project
~/projects/algorithms/luhn-verifier
Credit Card Verifier

C program that validates credit card numbers using the Luhn algorithm.

GitHub Project
~/projects/java/food-pyramid
Food Pyramid Simulator

Java program simulating a food pyramid/food chain ecosystem.

GitHub Project

Latest Achievements & Recognition

~/bmo/aml-pipeline
AML Case-Investigation Pipeline

Reduced a 3–4 hour manual AML investigation workflow to 1–2 minutes (~99% faster), with management projecting ~$2M in savings over two years.

Bank of Montreal

~/projects/llm/pi-mtplx
pi-mtplx Launch

Published a zero-configuration local LLM runtime for Pi, reaching 1,600 downloads in its first 2 weeks.

Personal Project (pi-mtplx)

~/systems/hpc-scaling
HPC & Distributed Systems

Enabled 100% collision-free horizontal scaling on the SeaWulf HPC cluster using dynamic sandboxing and isolation.

Reliable Systems Lab

~/systems/infrastructure
Systems Infrastructure

Accelerated validation latency by 99.9% via a scalable AWS EC2 CI/CD pipeline with asynchronous integrity checks.

Mailgator

Contact

Feel free to reach out for collaborations or just a friendly chat.

Email Me +1 (934)-219-4495 LinkedIn GitHub