Problem
Each domain hands people a confident answer with the reasoning stripped out: job products optimize for volume, screeners lead with a sortable number, fitness onboarding hands you a BMI label.
Three products. Built alone, end to end.
A private job-search desk with a public demo, a research board scoring 1,600+ stocks on India’s National Stock Exchange, and a free fitness assessment built for India. I designed and built all three, and each one refuses to hand over an answer without its reasoning.
60-second brief · Three solo products
Every answer shows its reasoning
job-lab gives judgment, not more job links. market-lab never shows a score without its doubt. fit-lab shows its work on every recommendation. Three domains, one refusal to hand over false confidence.
No answer travels without the reason behind it.
Outcome
Three working products, designed and built alone. market-lab and fit-lab are public; job-lab is a private desk with a public demo. Each ships with automated tests and an accessibility pass.
Move
Put the reason beside the answer: a worth-your-hour call tied to the next action, every score paired with its objection, every claim next to its evidence and limits.
Proof
7+ job-source families normalized, 1,600+ NSE stocks scored, 270+ unit checks across fit-lab’s engines, and an accessibility pass on all three.
Scope
I owned all of it: data pipelines, scoring logic, interfaces, light and dark themes, accessibility, tests, and deployment. No team, so no claim of team leadership.
Chapter 01job-lab
I needed judgment, not more job links
I never ran out of job links. I ran out of hours: working out which roles were real, whether remote included India, what the application should say, and who to write to. So job-lab scores whether a role is worth your hour and ties that score to the next action.
7+
public job-source families normalized
Greenhouse, Lever, Ashby, Workable, SmartRecruiters, Workday, and aggregators, cleaned into one consistent job object.
Read the full case study Open the job-lab demo (opens in a new tab)
The real desk is private and stays behind a login. The public demo runs on fictional data, so you can try every screen.
The first map
- Decorative geography
- Unclear political boundary
- Remote roles plotted as if they had a city
- No link to the ranked list
The map that works
- Accepted political map context
- India first, because that is the market
- Remote work counted separately
- Dots, city cards, and jobs all agree
Chapter 02market-lab
Every score has to show its doubt
A stock screener is dangerous when it looks cleaner than the market it describes. market-lab scores the NSE, and every row carries the reason to be skeptical: the score, its risk, its liquidity, the case for, and the case against.
1,600+
NSE stocks scored every trading day, from all 2,078 listed
A scheduled job fetches the data, computes the scores, verifies the output, and publishes static files. No runtime server.
Read the full case study Open market-lab (opens in a new tab)
The public board publishes derived scores, ranks and plain-language reasons only. The raw price, volume and deal series stay private, and it gives no horizon or position-sizing advice. An independent project, not affiliated with or endorsed by NSE, BSE or NSE Indices.
A typical screener
- Leads with a sortable number
- Treats liquidity as secondary
- Hides the counter-case in extra columns
- Leaves you to guess the time horizon
market-lab
- Pairs every score with its objection
- Shows order impact beside turnover
- Writes the argument in plain language
- Puts the horizon in the row
Chapter 03fit-lab
Every recommendation shows its work
Most fitness onboarding hands you a BMI label and turns a few preferences into a confident-looking plan. fit-lab asks only for measurements that change the answer, screens for reasons a generic plan could be unsafe, checks whether your goal and timeline agree, and attaches the evidence and the limits.
270+
unit checks across the deterministic product engines
Evidence, dose, goal, screening, capacity, and schedule. On top, one rule: no verified guide, no shipped exercise.
The shortcut
- BMI as the whole assessment
- Let the user pick muscles
- A seeded slider counts as an answer
- Caveats hidden in generic fine print
The fit-lab rule
- Use only measurements that change the read
- Prescribe complete movement coverage
- Require a real answer or interaction
- Put evidence and limits next to the claim
What AI did. What I did.
I design the system and direct AI agents to build it. I decide what it does, how it fails and what ships, and I review and test every change before it goes live. On all three, Claude, an AI coding agent, wrote the code to my direction.
job-lab
- What AI did
- Wrote the code that cleans 7+ public job-source families into one job object.
- What I did
- Decided it scores whether a role is worth your hour, and ties that score to the next action.
market-lab
- What AI did
- Wrote the code that scores the NSE.
- What I did
- Made every score carry its objection: its risk, its liquidity, and the case against.
fit-lab
- What AI did
- Wrote the code behind the onboarding and the plan.
- What I did
- Asked only for measurements that change the answer, and attached the evidence and the limits to every recommendation.
Why it still took me
Claude can write a stock screener. It will not decide on its own that every score has to show its doubt. That rule is the product.
What this proves, and what it doesn’t
I built all three alone, so the evidence is ownership end to end. Shipped is not the same as impact. Tests prove the products work. Only real use over time proves they help.
job-lab
What it proves
How I own a product end to end: the trade-offs, the implementation, and a working deployed system.
What it doesn’t
Team leadership. And it is too early for placement or conversion numbers.
market-lab
What it proves
Ownership end to end: model rules, data boundaries, responsive interaction, accessibility, recovery, tests, and deployment.
What it doesn’t
Team leadership, or investment returns. Judge it on research quality instead.
fit-lab
What it proves
The assessment logic, evidence links, body assets, anatomy coverage, responsive flow, themes, and tested accessibility paths work as designed.
What it doesn’t
It doesn’t diagnose or replace a clinician, and can’t prove health outcomes. No usage analytics yet, so no completion rates.
Three products. One rule.