Pass 1 of 3 Overview Who I am, and two results 1 minute

Quick View · the one-minute introduction

Ram Tewari

Purdue University · B.S. Industrial Engineering · Minor in Computer Engineering · Junior · Expected May 2028

I build AI systems with evidence behind their outputs and rules around their actions.

I’m Ram, an Industrial Engineering junior at Purdue. I work across code retrieval, AI agents and robotics. My focus is the whole system: how information gets in, how decisions are checked, and how the result is measured.

Creative about the question. Methodical about the proof.

Résumé (PDF)Contact Ram

Two results

  1. PHAROS AI code retrieval infrastructure

    Built commit-verified context for coding agents. Reported held-out retrieval ranking improved from 0.394 to 0.697 nDCG@10.

    Reported 101 questions across two held-out sets; reported, not independently reproduced here. Repository memory v1 failed its primary gates: 55% audited precision.

  2. Ultraviolette Applied ML engineering

    Built retrieval and deterministic checks for an EV showroom coaching pilot. A 40-session A/B test cut prompt tokens 24% while retaining detection of all 11 planted errors.

    Owner confirmed Résumé result; a narrow experiment, not a general guarantee.

Pass 2 of 3 How it works Four stories: the problem, what I owned, the decision, the result 3 minutes

What I would bring to your team

PHAROS AI code retrieval infrastructure · 2025 to now

Coding agents need context tied to the version of the repository they are working on. I built PHAROS: Tree-sitter indexing of symbols, imports and calls, retrieval filtered to one repository, and results served over REST and native MCP. The decision that shapes it: returned code is checked against the Git blob it was indexed from, instead of trusting a search result by default. That proves integrity, not correctness.

Reported Held-out nDCG@10 rose 0.394 → 0.697 and 0.463 → 0.726, across two held-out sets totaling 101 questions; single annotator, not reproduced here. Repository memory v1 failed its primary gates: 55% audited precision, primary source-file recall 0.827 → 0.823. Beside that failure, file-selection precision rose 63.4% → 77.5% in an 80-task test Owner confirmed.

How it picks a search mode

Auto mode prefers dense retrieval, uses hybrid with hints and falls back to keyword with a warning. An explicitly requested mode that is unavailable fails. Sparse (SPLADE) retrieval is implemented in an earlier search module; the audited auto mode does not route to it.

Ultraviolette Automotive Machine Learning Engineer Intern · summer 2026

Noisy showroom conversations had to become coaching feedback grounded in what was said and in versioned product specifications. In an EV showroom coaching pilot I benchmarked on-device speech processing, built analysis that links each finding to transcript evidence, and built the evaluation. The decision: code with numeric tolerance bounds decides whether a specification claim is correct; the model interprets the conversation.

Owner confirmed 96.5% speaker attribution on 50 hand-annotated recordings. A 100-session evaluation harness. Separately, a 40-session A/B test: 24% fewer prompt tokens, 11 of 11 planted errors still caught. Résumé results, not reproduced here.

Valerie / CognitiveOS Native AI assistant with validated actions · The Keeper’s Study

CognitiveOS is my native macOS and iOS app, and Valerie is its assistant. A model proposes one typed move; deterministic code checks it against the state it was issued and the bounds of its arguments before anything runs. This site’s guide uses the same kind of boundary, acting only on these pages. Try to break it.

Magna International Software Intern · warehouse robotics · summer 2025

An operator needed to be able to step in while a warehouse robot moved. I built the React interface (camera feeds, keyboard and joystick) and the Python/FastAPI backend that sends motion commands over SSH, with operator-initiated hand-offs between autonomous and manual control. Command latency: 58 ms keyboard, 62 ms joystick Owner confirmed.

Pass 3 of 3 Deep dive How I work, skills by the work, more work and next steps as long as you like

How I work

  1. Reframe the problem. Start from a better question than the one on the ticket. PHAROS
  2. Decide what would prove it. Write the bar down before building, so the result can’t move it. Ultraviolette
  3. Make it something you can touch. Build the smallest version a person can actually try. Magna
  4. Measure, and keep what fails. Report the result against the bar, and publish the misses beside the wins. PHAROS

Skills, by the work

Retrieval and evaluation PHAROS, Ultraviolette
Python, FastAPI, Tree-sitter, vector search, RAG, retrieval evaluation (nDCG), MCP
Speech and models Ultraviolette, R&Bot
PyTorch, Transformers, Whisper, pyannote, edge AI
Apps and agents Valerie, Magna
SwiftUI, TypeScript, React, typed action validation
Systems and languages Across the work
PostgreSQL, pgvector, Redis, Docker, Git, Linux, NVIDIA Jetson; C++, C, SQL, JavaScript, Java, SystemVerilog

More work

  • R&Bot: a 33M-parameter transformer trained from scratch for text generation, four epochs on about 50M tokens, with section tokens for verse, chorus and bridge Owner confirmed. Source and checkpoint are private.
  • BABEL: A Python/React prototype that turns prose into structured visual-scenario blocks. Source verified
  • Earlier: Yulu Bikes, Mechanical Engineering Intern, 2023: chassis design, CAD and prototyping.

Next steps

Résumé (PDF)tewari10@purdue.eduGitHubLinkedIn