Prashant Shah

Selected Work

Stealth Startup · 2026

Atlas LMS

Revolutionizing learning. More soon.

HackRamapo Winner · 2026

Canvas Agent

Next.js 15React 19 PostgreSQLpgvector Vercel AI SDKTailwind CSS

The agent combines persistent memory, tool calling, and semantic search over Canvas data to answer questions across courses, track grades and deadlines, and surface what a student should work on next.

Students leave the LMS to make sense of the course because it stores records without helping them reason across deadlines, grades, assignments, and study priorities.

I built an AI study partner that works directly from a student’s Canvas data, remembers context across conversations, and can use tools instead of only generating text.

Winner of HackRamapo 2026.

Personal · 2024

Moodie

Movie recommendations ignore how you're feeling. On-device facial emotion detection mapped to genre preferences via hybrid scoring. 69 tests covering auth, SQL injection, and recommendation accuracy.

ReactNode.jsPostgreSQL

View Live → View on GitHub →

Personal · 2026

Hermes JobApps

I built a job-search command center around Hermes Agent. It discovers and evaluates roles, tailors LaTeX materials, organizes evidence and contacts, and tracks applications, outreach, and follow-ups in SQLite so the work survives beyond a chat session.

PythonSQLiteHermes AgentTool UseLaTeX

View on GitHub →

I built the dashboard, SQLite data model, Hermes plugin and tool layer, streaming sessions, material-prep pipeline, provenance records, and human-controlled boundaries for uploads, sends, and submissions.

Trimble · 2025

Configuration Management Platform

Config conflicts across a distributed broker architecture cost engineering teams hours. I built a production platform in C#/.NET and Azure centralizing service settings across messaging, event handling, and dispatch.

C#.NETAzureBlazor

Auth proxy using Azure Workload Identity Federation for role-based access. REST APIs integrating with TMW Suite. Built and shipped to production, not a prototype.

Novartis · 2024

Project Proposal Automation

Proposals took 2 days through a manual approval pipeline. I built a React/TypeScript frontend with serverless Azure Functions, integrated with SharePoint. Real-time dashboards for budget and team tracking.

ReactTypeScriptAzure FunctionsSPFx

Ramapo DMC - Center for Food Action · 2025–Present

Food Pantry Data Infrastructure

I am overhauling the data path from donation intake through pricing, inventory, review, PantrySoft handoff, reconciliation, dashboards, and recurring reports. The organization reaches about 40,000 people annually across 6 pantries. Featured on News 12 NJ.

PythonFlaskPostgreSQL

Data Science Capstone · 2026

DeepScenario Collision Predictor

Full-fidelity autonomous-driving simulations are expensive. I built a screening model that ranks 33,366 scenarios from initialization-only features, then tested how well collision risk transfers across random, greedy, and reinforcement-learning scenario generators.

PythonXGBoostGroupKFoldDistribution Shift

View on GitHub →

Research

Resolving Intent-Content Conflict in Retrieval

When someone asks for "a happy movie about war," intent and content semantics disagree. Most systems optimize one signal. This work asks whether you can separate topical match from affective fit and resolve the conflict explicitly.

How: KRAG on MovieLens 25M: a heterogeneous knowledge graph with User, Movie, and Emotion nodes and weighted EVOKES edges. A Graph Transformer encoder is aligned to Sentence-BERT so graph and text scores fuse coherently; ranking uses semantic + graph relevance minus a tunable affective displacement penalty.

What we found: Dissonance (topic vs. emotion conflict) is the hard case, and where evidence concentrates: larger gains vs. baselines there than under agreement, including a large paired effect on affective displacement when the penalty is on (d_z = −1.51). Edge-removal tests show emotion edges are causally necessary; necessity is ~2.3× higher under dissonance than agreement, so the graph matters most when signals clash.

KRAG Paper accepted to NGEN-AI 2026. Received the Best Research Award at Ramapo College. Presented at the Ramapo DMC Fair and Scholar’s Day.

KRAG on GitHub →

1,600

synthetic queries (agreement vs. dissonance)

−1.51

large d_z paired effect: affective displacement error drops in conflict when the penalty is on

Causal

perturbing dominant EVOKES edges drops fused scores; baselines that ignore the graph show zero necessity by design

~2.3×

higher edge-necessity (FNS) in dissonance vs. agreement

Experience

Sep 2025 — Now

Data Engineer · Ramapo DMC - Center for Food Action · Contract

Overhauling the data path from intake through pricing, inventory, reconciliation, PantrySoft handoff, dashboards, and reporting.

Featured on News 12 NJ

May — Aug 2025

Software Engineering Intern · Trimble

Configuration management platform shipped to production.

C# · .NET · Azure

May 2025 — May 2026

Research Assistant · Ramapo College

Built and evaluated a Vertex AI retrieval pipeline over MovieLens 25M for cases where topical relevance and user intent disagree.

Best Research Award

Jun — Aug 2024

Software Engineering Intern, Business Analytics · Novartis

A recurring 2-day reporting workload became refreshable views across 100+ initiatives.

PostgreSQL · Power BI

Aug 2023 — Jan 2026

Data Analyst · Ramapo College

Built Python and Quarto reporting workflows across 6,000+ student records, surveys, attendance, performance, and retention data.

Student Success

Sep 2022 — May 2023

Undergraduate Researcher · Ramapo College

Mapped Self-Determination Theory and Bloom’s Taxonomy to feedback, game mechanics, and an introductory IT course deployment plan.

SAM Conference · Top Poster

Class of 2026

B.S. Computer Science & Data Science · Ramapo College of NJ

Mathematics minor.

GPA 3.8

Languages Python · C# · TypeScript · Java · C/C++ · SQL · Bash

Backend & Data PostgreSQL · .NET · Flask · FastAPI · Node.js · pgvector · SQLAlchemy · ETL

Cloud & Ops Azure · Azure Functions · AWS · Docker · Linux · GitHub Actions · CI/CD

AI & Product Tool-calling agents · RAG · embeddings · React · Next.js · Vercel AI SDK

Practices pytest · Playwright · schema design · validation · code review

About

I grew up speaking three languages and moving between worlds. That shapes how I see problems: I'm always translating, always aware there's more than one way to frame something.

I came to computer science because software is one of the few tools that actually scales. At Ramapo I studied CS and Data Science and worked as a Data Analyst, building reporting workflows over student records, surveys, attendance, and performance data. I also managed 17 resident assistants and a residential community of 200+ students.

At Trimble, Novartis, and Ramapo DMC - Center for Food Action, I kept returning to the same problem: useful software starts with understanding how data moves, where decisions break, and what the person using the system needs to trust. My research applies that question to retrieval systems and learning technology.

I am building Atlas LMS toward a startup while continuing the Ramapo DMC - Center for Food Action infrastructure overhaul. I am also looking for full-time Software Engineer and Data Engineer roles where I can own real systems.

Canstruction: 1st place, building structures from canned food

Best Research Award, Ramapo College

Presidential Scholar

ICPC competitor

Now Updated July 2026

Building

Building Atlas LMS as a startup. At Ramapo DMC - Center for Food Action, overhauling the path from intake and inventory through reconciliation, dashboards, and reporting.

Researching

The KRAG Paper was accepted to NGEN-AI 2026. I am continuing the research thread on what retrieval systems do when topical similarity and a user’s actual intent point in different directions.

Learning

Understanding harnesses, building AI tools, and product scope. Also photography.

Looking for

Full-time SWE / Data Engineer / AI Engineer roles. Recently graduated in May 2026.

Open to

Conversations with engineers, founders, and researchers working on data systems, learning products, or AI that has to operate inside a real workflow.

Contact

shahp0021@gmail.com