We're hiring

We don't hire for credentials.
We hire for obsession.

We're building a cognitive classification engine trained on a 2,048-type personality system that has never been formally validated or computationally modelled. Two founding roles. Both first hires.

Open roles

Choose a role

Role 01 · Research
Founding Psychologist / Researcher
₹12–18 LPA + Equity · Bangalore
Design and run the first formal validation studies on OPS. Inter-rater reliability, construct validity, and the multi-year research programme that makes this system scientifically credible.
Read more & apply →
Role 02 · Engineering
Founding Engineer
₹25–35 LPA + Equity · Bangalore
Build the classification engine. Extract features from video/audio/text of typed subjects, build models that predict type codes, and ship it as a product. ML + systems, end to end.
Read more & apply →
01

The problem

The core question is simple: does OPS measure what it claims to measure?

But answering that question requires you to go deep. This is not a role where you show up, run a standard study protocol, and write a paper.

Completely understand how the system works
OPS has 2,048 types across 9 binary coins and 4 social types. You need to learn it from the inside — study how typing works, engage with the practitioner community, understand the logic. Not as an outside evaluator. As someone who gets it.
Go into your own unconscious and biases
OPS makes a specific claim: people cannot accurately see their own cognitive patterns. A researcher who doesn’t engage with that claim personally won’t understand what they’re measuring.
Think from first principles, not jargon
You will need to question the foundations of both OPS and academic psychology. If your instinct is to accept what you’ve been taught as settled truth, this isn’t the right role.
Design validation from scratch
There is no existing academic literature on OPS and no template to follow. Phase 1 is an inter-rater reliability study — double-blind typing sessions across 500+ participants, computing Cohen’s kappa for each of the 9 coins.
Research how to eventually remove human involvement
The long-term product is a classification engine, not an army of trained typologists. You’ll think deeply about what the typologists are observing and whether those signals can be captured computationally.
Conduct many studies, not just one
Phase 1 is inter-rater reliability. Phase 2 is construct validity. Phase 3 is predictive validity. Phase 4 is heritability. This is a multi-year research programme.
02

What you'll actually do

Year 1Learn the system +
Inter-rater reliability
Immerse yourself in OPS — study the framework, get typed yourself, engage with practitioners
Design the Phase 1 study: double-blind typing sessions, participant recruitment, consent procedures, ethics documentation
Operationalise practitioner terminology into precise behavioural anchors suitable for academic scrutiny
Coordinate with OPS-trained typologists to run typing sessions
Analyse data and co-author the first peer-reviewed paper on OPS
OngoingBuild the research
programme
Shape and execute Phases 2–4: construct validity, predictive validity, heritability studies
Research computational approaches to typing — what can be captured without human observers?
Build scientific credibility — conference presentations, academic partnerships, pre-registration
Translate research findings into product and investor-facing narratives
03

Who you are

We care less about your degree and more about how your mind works.

Mindset
You think from first principles
You question frameworks — including the ones you were trained on. You don’t treat academic consensus as gospel.
Obsession
Deep interest in the human mind
Not as an abstract academic topic, but as something you’ve spent years thinking about. Why people behave the way they do, why self-knowledge is so hard.
Craft
You know how to design and run studies
Inter-rater reliability, Cohen’s kappa, construct validity, factor analysis — things you’ve actually done, not just read about.
Rigour
Intellectually honest
If the data says OPS doesn’t work, we need to know that. Truth over confirmation.
Commitment
Long-term
This is a multi-year research programme. The person who starts this work should be the person who finishes it.
Bonus
Strong advantages, not required
Jungian typology or personality type systems. Observational coding. Psychology × technology. Startup experience. R or Python.
04

Compensation

₹12–18 LPA + Equity
Negotiable based on mutual excitement
If you’re the right person and we both know it, we’ll figure out the number. Equity participation is on the table.
Show us how you think.
No cover letter templates. No generic applications. We read everything.

    01

    The problem

    Trained OPS typologists watch a person and classify them across 9 binary coins. They’re seeing something — patterns in speech, body language, word choice, facial expressions, vocal tonality, response patterns.

    The question is: what exactly are they seeing, and can a model learn to see it too?

    This is a multimodal classification problem. Video + audio + language data for hundreds of typed subjects with known type codes. Figure out which signals predict each coin — and eventually build a system that classifies without human observers.

    There’s no Kaggle dataset, no benchmark, no existing literature. You’ll define the feature space, the architecture, and the evaluation criteria from scratch.

    Real data exists today
    500–2,000+ officially typed subjects with video recordings, type codes, and a full practitioner-built database. No one has touched this with ML tools.
    A scrappy MVP already works
    The founder has built a system calling Claude and Gemini APIs that analyses influencer content and matches it to brand campaigns based on OPS type data. It’s early, but it works well enough to test ideas against. You’ll experiment with this and other approaches.
    02

    What you'll actually do

    Phase 1Feature extraction
    & exploration
    Evaluate and iterate on the existing MVP — understand what works, what doesn’t, what needs rebuilding
    Build the data pipeline — ingest video recordings, extract audio, generate transcripts, structure against the typed subject database
    Extract features across modalities: facial action units, vocal prosody, speech rate, word frequency, body movement, gestures
    Explore which features correlate with which coins
    Work closely with the founder (7 years of OPS knowledge) to translate what typologists observe into computational hypotheses
    Phase 2Classification
    models
    Build classifiers for individual coins — start with strongest signals
    Experiment with architectures: coin independence, interaction effects, LLM-based vs. custom models
    Evaluate confidence and failure modes — where does the model disagree with typologists?
    Feed findings back to the research track — model performance is evidence for or against coin validity
    Phase 3Product
    Build toward a classification API — input: video/audio/text, output: structured type code with confidence scores
    Figure out minimum viable input — 30 minutes of video or 5? Audio only? Text only?
    Inference pipeline, latency, scalability
    03

    Who you are

    Builder
    You build things end-to-end
    Train a model Monday, write the pipeline Wednesday, deploy the API Friday. You are the team. Python, PyTorch, and enough systems engineering to ship real products.
    ML
    Experience with unstructured data
    You’ve worked with video, audio, or NLP — ideally more than one. You extract meaningful features from messy, real-world media.
    Ambiguity
    No benchmark, no leaderboard
    No existing baseline. You define what success looks like. If you need a well-scoped problem handed to you, this isn’t it.
    Thinking
    First principles, not architecture tourism
    You think about why a particular approach should work for this problem. You reason about the domain, not just the math.
    Curiosity
    Interested in the domain
    You don’t have to know OPS. But “why do humans behave differently” should be an interesting question to you, not a boring one.
    Commitment
    Long-term
    You’re not joining an engineering team — you’re building it. Not a 6-month contractor.

    Strong advantages (not required): affective computing, computational social science, video understanding models, production ML APIs, infra (AWS/GCP, Docker), personality psychology familiarity, startup experience.

    04

    Compensation

    ₹25–35 LPA + Equity
    Negotiable based on mutual excitement
    If you’re the right person and we both know it, we’ll figure out the number. Equity participation is on the table.
    05

    Why this role is unusual

    Most engineering roles: optimise this ad model, build this CRUD app, fine-tune this LLM. Well-scoped, incremental, someone already defined the problem.

    This is different. First engineer at a company building a classification system for something never computationally modelled before. The data is real. The feature space is undefined. The architecture is yours.

    You’re not joining an engineering team. You’re building it.

    Show us what you build.
    No cover letter templates. No generic applications. We read everything.

      Questions? abhas@mahakram.in ← Back to main page
      Mahakram Labs
      Research · Validation · Infrastructure