Portrait of John Cedric Alora

Machine learning research and engineering

John Cedric Alora

Computer science senior at Cal State Fullerton and Cal-Bridge Scholar. I work on retrieval for AI systems: embeddings, vector search, and knowing when the answer can be trusted.

  • Cal State Fullerton
  • Cal-Bridge Scholar
  • UC Merced
  • Project RAISE
  • DSML Club President
  • AIxDATA Datathon, 1st place
  • Bakersfield College
John in a suit on the Bakersfield College campus

I'm John. I started college in 2019, could not pay for it, and worked my way back through warehouses and door-to-door sales. Now I'm a computer science senior, a Cal-Bridge Scholar, and an undergraduate researcher working on vector search and machine learning.

Now
CS senior at Cal State Fullerton, graduating May 2027
Interests
Retrieval-augmented generation, vector search, ML evaluation
Next
PhD programs and ML roles, Fall 2027

Selected work

John holding his prize on stage at the AIxDATA Datathon

1st place, AIxDATA 2026 Datathon

Caller Companion

A briefing tool that gives outbound sales agents a probability for each call before they dial, with plain-language talking points. Built solo in 24 hours against teams of 4 to 6. LightGBM on 41,188 call records, with the leaky feature removed.

  • Python
  • LightGBM
  • FastAPI
  • Next.js
John with his advisor, his mentor, and the lab group at UC Merced

Research, UC Merced

Filtered vector search

A summer in Dr. Florin Rusu's lab on the search layer under retrieval systems: benchmarking FAISS and HNSW under filters, and building test queries hard enough to tell good plans from bad ones. Presented at the UC Merced summer symposium and the Cal-Bridge Research Symposium.

  • Python
  • FAISS
  • HNSW

2026 to 2027, team of five

Industry-sponsored machine learning capstone

Selected from my resume and experience to work year-round under an industry mentor. A classifier that has to work under noisy real-world conditions and explain its predictions.

  • Machine learning
  • Explainable AI
A team meeting at the Arc Energy Solutions office

Arc Energy Solutions

Arc Portal

One dashboard over the sales and operations CRMs, with ad and lead data pulled in through APIs. Includes an assistant that answers from the company Drive and keeps each salesman out of other people's deals.

  • Next.js
  • Supabase
  • PostgreSQL

Research paper and poster, with Dr. Kanika Sood

Does a model's confidence survive new people?

On people they never trained on, KNN and a decision tree land two points apart in accuracy. Their calibration error is three times apart.

Accuracy

KNN87.3%
Tree85.1%

Calibration error

KNN0.047
Tree0.149

Reinforcement learning for Pokémon Red

Custom reward rules on an open-source environment, with an ablation study of agent behavior with and without them.

PyTorch, Stable-Baselines3

Pulse

My first hackathon, hosted by IBM at Cal State LA. A health app that turns doctor visit documents and symptoms into a recommendation.

IBM watsonx, multi-agent

Kingler

A personal AI assistant running on a local machine: daily briefings, alerts, and coursework organization.

Local LLM, Supabase

live vector search

vectors
1,700
k
10
graph hops
0
brute-force scan
0.000 ms
k-th neighbor
0.0 px
filter
none

Research

I'm interested in retrieval-augmented generation (RAG) and the whole path under it: text becomes embeddings, embeddings become a graph in many dimensions, and a search walks that graph to find what a model needs. This is that search, running live. Move your cursor. The orange path hops point to point toward you.

How an AI system finds an answer, and where I've worked on it

01 / 04

Embed

Text, images, and records become vectors: long lists of numbers where similar things land close together.

Where I've done this

  • Arc Chat, Arc Energy SolutionsBuilt an assistant that pulls answers from the company's Google Drive, stored in Postgres with pgvector.

02 / 04

Index

Millions of vectors can't be checked one at a time. They get linked into a graph that a search can walk. HNSW is the common one.

Where I've done this

  • UC Merced, Dr. Florin Rusu's labBenchmarked FAISS and HNSW indexes in Python.

03 / 04

Retrieve

A query walks the graph to its nearest neighbors. Add a filter, like "only this customer's files", and the search gets much harder.

Where I've done this

  • UC MercedWorked on filtered vector search: comparing pre-filtering with post-filtering, and building harder test queries.
  • Arc ChatScoped retrieval per user, so a salesman can't pull up someone else's deals.

04 / 04

Trust

The model answers from what it retrieved. Then the question I care about most: is it right, and was the test hard enough to tell?

Where I've done this

  • Cal State Fullerton, with Dr. Kanika SoodStudied whether a model's confidence holds up on people it never trained on.
  • Caller CompanionFound the feature that leaked the answer and removed it before reporting results.

Research, work, and leadership

  1. Sep 2026 to now

    Undergraduate Research Assistant

    Dr. Kanika Sood's lab, Cal State Fullerton

    Read and score student research papers through a conference-style review process, and support ongoing ML research.

  2. Sep 2026 to May 2027

    Machine Learning Capstone

    Industry mentored, CPSC 490 at Cal State Fullerton

    Selected from my resume and experience to work year-round under an industry mentor. Under NDA.

  3. 2026 to 2027

    President, Data Science and Machine Learning Club

    Cal State Fullerton

    More than fifteen active members. Putting students in front of research they would not otherwise see, and organizing a datathon this year.

  4. May to Jul 2026

    Research Intern

    UC Merced (SURI), advised by Dr. Florin Rusu, mentored by Abylay Amanbayev

    Filtered vector search with FAISS and HNSW. Built harder test queries for the lab's benchmark. Presented at the Cal-Bridge Research Symposium. Co-author on a paper in preparation.

  5. Jun to Aug 2026

    Student Research Mentor

    Project RAISE, Cal State Fullerton, under Dr. Kanika Sood

    Mentored four community college students through their own ML projects and built the curriculum they worked from.

  6. Feb 2024 to Nov 2026

    CRM Automation Lead

    Arc Energy Solutions, Bakersfield and remote

    Started as a door-to-door solar sales rep. Moved the team to a digital lead pipeline, raised conversion about 30%, and owned the CRM platform.

  7. Summer 2025

    Research Workshop Participant

    CIC-PCUBED Transfer Pathways, Cal State Fullerton

    First research exposure: regression and clustering on a public dataset about social media use and productivity.

Education

California State University, Fullerton
B.S. Computer Science, AI/ML concentration. Minor in Data Science.
Expected May 2027

Bakersfield College
A.S. Computer Science and Liberal Arts
May 2025

Skills

Languages
Python, SQL, JavaScript, C++

ML and data
PyTorch, scikit-learn, LightGBM, pandas, NumPy, FAISS / HNSW

Systems
PostgreSQL, Supabase, REST APIs, Git, Linux / WSL

Recognition

Cal-Bridge Scholar, Cohort 12
Statewide PhD preparation program for CSU STEM undergraduates.

1st place, AIxDATA 2026 Datathon
CSU Long Beach AI Club.

Where I've been

2019 to 2025

Bakersfield College

I started here, left to work, came back, and finished.

2024 to 2026

Arc Energy Solutions

From knocking on doors to running the systems the sales team runs on.

2025 to now

Cal State Fullerton

Transfer student, CS senior, and president of the Data Science and Machine Learning Club.

Summer 2026

UC Merced

My first research lab, and the work I'm most proud of.

Now

Cal-Bridge

Cohort 12. The program that showed me a PhD was an option.

John in cap and gown at his Bakersfield College graduationA team meeting at the Arc Energy Solutions officeThe Titans sign outside the Cal State Fullerton student unionJohn with his advisor, his mentor, and the lab group at UC MercedJohn with the Cal-Bridge director at the fall conference