Aditya Rathi

Machine learning engineer and tech lead for NLP and LLM systems on large-scale production data

First engineering hire at Burning Glass Institute, New York

I lead the institute's models on education (schools, degrees, majors), location, wages and credentials, and its record matching. With a small team, I build the Credential Value Index pipeline, which turns hundreds of millions of credential records into a public index states use to fund programs.

GitHubLinkedInResume (PDF)

One fictional credential record through the pipelineThe raw record holds only a credential name, "adv. mfg tech cert (lvl 1)", and an organization, "LARKSPUR TECH COLL - WRKFRC ED". It becomes one approved catalog row: provider Larkspur Technical College, credential Advanced Manufacturing Technician, field Manufacturing technology, skills CNC machining, Blueprint reading and Quality inspection.adv. mfg tech cert (lvl 1)LARKSPUR TECH COLL - WRKFRC EDas it arrivesProviderLarkspur Technical CollegeCredentialAdvanced Manufacturing TechnicianFieldManufacturing technologySkillsCNC machining · Blueprint reading· Quality inspectionone catalog row, approved
This record is invented and the steps are simplified. It follows one record through the pipeline.
  1. A raw record arrives

    It holds only a credential name and an organization. It comes from one of about 20 sources, each written its own way.

  2. Provider

    The organization is tied to one record per real provider. The July 2026 release lists 1 million of them.

  3. Credential

    The name is matched to a clean reference list and given a field of study.

  4. Skills

    The credential is linked to the skills it teaches.

  5. One catalog row

    A person approves each release before it is published.

Selected systems

Earlier

At Sophos AI Research in 2022, I added learned text features to a security alert-prioritization system, and they shipped to production.

BrainAI was a 2022 course project that identified people from EEG and ECG signals, with an F1 of 77 percent (vs 0.5% chance level for 200 subjects).

Timeline

  1. Jun 2023 to presentMachine Learning Engineer, Burning Glass Institute
  2. May 2023M.S.E. Data Science; B.S.E. Electrical Engineering, summa cum laude, University of Pennsylvania
  3. Jun 2022 to Dec 2022Data Science Intern, Sophos AI Research

Contact