Rohan Singh

Rohan Singh

About me

I work where clinical context meets AI systems.

My academics cover healthcare and its technology. A B.Pharm from DIPSAR gave me a foundation in pharmacy and regulation. An M.Tech in Clinical Engineering from IIT Madras, a programme run jointly with CMC Vellore and SCTIMST, put me inside real clinical settings. Along the way I did projects in product development, risk assessment, biomaterials and app development.

At Outlier, I'm part of the INQA program and work mostly on the quality side of frontier model research across coding, fintech and healthtech, supporting RLHF and SFT workflows. The common thread is judging correctness in high-stakes domains.

I'm passionate about MedTech product development and healthcare AI, where clinical grounding and AI fluency both matter.

Education

  • M.Tech, Clinical EngineeringIIT Madras, run jointly with CMC Vellore and SCTIMST. 2023 to 2025, CGPA 8.38
  • Bachelor of PharmacyDelhi Institute of Pharmaceutical Sciences and Research. 2019 to 2023, CGPA 7.68

Registration and writing

  • Registered PharmacistDelhi Pharmacy Council, 2026
  • Advances in piezoelectric biopolymer materials for high-performance biomedical sensing and actuationPublication

Courses and volunteering

  • McKinsey Forward ProgramMcKinsey and Company
  • Python for Data Science, AI and DevelopmentIBM on Coursera
  • Google AI EssentialsGoogle
  • Fabricate 2025, IIT MadrasLogistics and speaker coordination

Evaluation

Rubric design, failure analysis, LLM evaluation, SFT and RLHF workflows, golden answers

Data

Python (Pandas, NumPy, scikit-learn), SQL, Power BI, Excel modelling

Building

LangChain, FAISS, Streamlit, Docker, Kotlin, Git, Claude Code, Cursor

Clinical and product

Needs finding, requirements, user stories, Figma, ISO 14971

I test how AI handles healthcare work, and find where it goes wrong.

Registered pharmacist with an M.Tech in Clinical Engineering from IIT Madras. I grade model answers against written rubrics, build evaluation tasks with known answers, and turn what clinicians tell me into specifications.

Question

Which ICU infusion pumps are overdue for preventive maintenance?

Model answer

Three ICU pumps are overdue: IP-07, IP-12 and IP-19. Factual

IP-12 was last serviced on 2 March 2026 and the interval is 12 months, so it fell due in September. Reasoning

I confirmed this in the maintenance log and the asset register. Tool use

All three pumps are assigned to the ICU.

  1. Lists all four overdue ICU pumps
  2. Applies the 12-month service interval
  3. Works out each due date correctly
  4. Cites only the sources it opened
  5. Keeps to pumps in the ICU
0 of 5 criteria met
A made-up example of how I review a model answer: each claim checked against the source data, each failure given a type.

Experience

  1. AI Quality Assurance Contributor

    Outlier (Scale AI), remote

    • Reviewed model output on more than 300 tasks in finance, coding, STEM and healthtech, checking facts, reasoning and adherence to guidelines.
    • Classify failures by type (factual, reasoning, tool use) against atomic and process or outcome rubrics, and write criteria that other reviewers can apply the same way.
    • Wrote multi-step tasks for a simulated accounting firm with 12 connected business systems, where the answer means tracing a transaction across systems.
    • Squad Lead on a supervised fine-tuning project, May to Jun 2026. Onboarded, trained and mentored 18 contributors through live sessions and reviews. All 18 kept quality ratings of 4.5 or higher.
  2. Healthcare Benchmark Task Author

    Caudal AI, freelance

    • Wrote expert-level tasks that test whether AI models can do real hospital work, across pharmacy, clinical engineering, facility design, hospital operations and health information management.
    • Each task came with realistic input data, a spreadsheet reference answer built on live formulas, and a weighted rubric used to score model output.
    • Built Python tooling that generates the input data and checks that every stated figure can be recomputed from the files.
  3. Power BI Intern

    Celebal Technologies, remote

    • Built a product performance and sustainability dashboard for two industrial product lines, with profit-score and quality KPIs, drill-throughs and a mobile layout, during an 8-week program on DAX, ETL and data modelling.
  4. Clinical Intern

    Sree Chitra Tirunal Institute for Medical Sciences and Technology, Thiruvananthapuram

    • Semester of coursework, lab work and clinical rotations in medical device technology management.
    • ISO 14971 risk analysis for a glucometer, the unmet needs project, and lab work on gene delivery and wound healing.
  5. Clinical Engineering Intern

    Christian Medical College, Vellore

    • Observed primary and community health centres to see how devices hold up in low-resource settings.
    • Built the low back pain tracking app and volunteered at the Rehab Mela.
  6. Hospital Pharmacy Intern

    Rajiv Gandhi Super Speciality Hospital, Delhi

    • Dispensing, record keeping, adverse drug reaction reporting and patient counselling. Flagged more than 10 invalid prescriptions.
  7. Earlier

    Youth Ambassador, Viral Fission, Nov 2021 to Jan 2023. Campus campaigns for brands including Spotify and Fastrack.

    Business Analyst, MedTourEasy, Mar 2021. Dashboards to track KPIs.

Work

Clinical projects and things I built to test an idea. Most of it sits where hospital practice meets data or software.

PayLens

Personal project, 2025

An assistant that answers questions about Indian payment fees, from card and UPI charges to gateway pricing. It retrieves from a knowledge base of more than 150 fee-rule chunks and adds live search when the base has no answer.

  • Wrote an evaluation harness that reruns 20 benchmark questions after every change and scores context coverage, faithfulness and relevancy.
  • Packaged with Docker and deployed on Hugging Face Spaces.

Python, LangChain, FAISS, Sentence Transformers, Streamlit, Docker

Device lifecycle and risk platform

Personal project, Jan to Apr 2025

A prototype for hospital biomedical teams. It scores risk across the device register and shows only what is overdue or due soon, so nobody has to scan the full list.

  • Mapped four users: biomedical engineer, nurse supervisor, pharmacist and hospital administrator.
  • Wireframed their flows in Figma and built the Power BI dashboard on simulated data, based on workflows I saw at CMC Vellore and SCTIMST.

Power BI, Excel, Figma

Power BI dashboard with device counts by location and status, an equipment table and technician workload
Maintenance dashboard, simulated data
Prototype walkthrough, 1 min 35 s

Vapour-responsive chitosan films

M.Tech thesis, IIT Madras, Dec 2024 to May 2025

Chitosan films that bend when exposed to water vapour, tested with and without a barium titanate filler, and a small switch that uses the bending to close a circuit.

  • Cut fabrication effort by 70% over repeated rounds of process changes.
  • Built a sensor prototype on an ESP32 that streams readings in real time.
  • Presented as a poster at Fabricate 2025, IIT Madras.

Solvent casting, materials testing, ESP32, Fusion 360

Grid of film strips bending in water vapour, one row per filler amount
Film response at four filler amounts
The switch lighting an LED, 3x speed

Unmet needs in hospitals

Clinical internship, SCTIMST, Aug to Nov 2024

I led a team of four through interviews and floor observation across hospital departments. We found and validated more than 30 unmet clinical needs, screened them with a weighted matrix, and wrote a technical specification for a wireless 12-lead ECG monitor.

Needs finding, stakeholder interviews, requirements writing

Low back pain tracking app

Clinical internship, CMC Vellore, Jan to Jun 2024

An Android app where a person logs back pain in the morning, afternoon and evening, next to sleep, medication, work and activity, so the pattern can be reviewed later.

  • Wrote the user stories and built it in Kotlin, with data sent to Google Sheets through Apps Script and kept offline in SQLite.
  • Installed it on 10 people's phones to test feasibility, bugs and performance.

Kotlin, Android Studio, Google Apps Script, SQLite

Daily survey screen with sleep rating, pain sliders and medication questions
Short term projects and lab work
EV operations analysis154,000+ GPS and battery records from 1,200+ vehicles. Clustered five operating zones and flagged about 20% low-activity users.2025
Glucometer risk analysisDevice characterisation and an ISO 14971 hazard analysis, with the Division of Regulatory Affairs, SCTIMST.2024
Nanoplexes for gene deliveryFormulated polymer-based nanoplexes for intracellular gene delivery, Biosurface Technology Lab, SCTIMST.2024
Wound-healing assayIn-vitro scratch assay on L929 cells to study the effect of ascorbic acid on wound closure and collagen.2024
Oscilloscope plotting toolStreamlit app that reads exported oscilloscope CSV files and plots voltage against time.2024Notes
AR distance measurementUnity and ARCore app that measures distance to an object from the live camera view. Virtual Reality Engineering course.2024
Neurorehabilitation device launch planBusiness model, go-to-market, regulatory route and financials for a medical device. Medical Device Innovation course.2024
Speech corpus annotationLabelled 20,000+ audio files against written guidelines for Bhashini, National Language Translation Mission, IIT Madras.2023

Contact

Open to roles in healthcare AI, evaluation and product. Email is the fastest way to reach me.