On April 6, 2026, a Madurai court sentenced nine policemen to death for the 2020 custodial deaths of Jayaraj and Bennix. A father and son walked into a Thoothukudi police station for a minor lockdown violation and never walked out.

Six years. Two deaths. One judgment.

On April 8, 2026, our SocEnt elective at ISDM began. The faculty was P. R. Ganapathy, who has mentored social entrepreneurs at Villgro, Stanford Seed, and Menterra. He opened with one line.

Fall in love with the problem, not the solution.

Most ventures fail not because of poor execution. They fail because they solved the wrong problem. Or solved a real problem for an imagined user. A well-defined problem is half the work.

Eight days later, our group had an NGO called ARAM, two products under it, and a working AI distress detection system. This is the field note on the seven sessions in between, because the work was not in the answer. It was in how many times we let the problem be redefined.

We were five. Shruti, Udaya, Charlie, Madhi, and me. The first WhatsApp thread was the usual scramble. Climate. Waste management. Sex education. Mental health. Garbage to glamour. Education in MP towns. Hydroelectric projects displacing communities in Himachal.

Everyone had something. Nobody had the thing.

Udaya wrote one line that changed the direction. I would suggest something we have experienced, because that is what they want.

The next afternoon, sitting at lunch, with the coordinator asking for the third time, I sent the group the Madurai judgment and said, “Let us look at custodial violence.” Police in India have no structured training in non-violent communication. Violence becomes the default tool of control. Punishment after the fact does not change behaviour before the incident.

Madhi: Noissss. Shruti: +1. Charlie: Lovely. Let's do this. Udaya: Wow. Yes.

The aligned thoughts became a group. We had a problem. We had no idea yet what we did not know.

The first tool was a discipline against premature solutioning.

Why does custodial violence happen? Pressure to close cases.

Why is there pressure? High caseloads.

Why are caseloads high? Chronic understaffing, fourteen-hour shifts, no Sunday off in three months.

By the third “why", we were no longer looking at police as villains. We were looking at a workforce broken by the system it serves. The constable who hits a suspect in lockup is also a man being crushed by his own institution.

That shift mattered. It is why our solution later refused to be a training program shaped like a punishment.

Ganapathy walked us through the four pathways. Government. NGO. Market-led. Social enterprise. Then the test: is there a paying customer? Is there value that can be priced? Is the problem persistent despite government and NGO efforts?

We rated honestly. Government and NGO insufficient: high. Market gap: high. Paying customer: low. Revenue model: low.

The math told us what we did not want to hear. Custodial violence has no market buyer. The state is the perpetrator and the gatekeeper.

Ganapathy’s written feedback came back: Social enterprise potential: LOW. Risk: becoming a grant-funded training provider, not a scalable enterprise.

Udaya posted in the document the next day: “We are not an SPO. We are an NGO.”

We had walked into the elective believing we were going to build a social enterprise. That was the entire point of the course. We were the only group that explicitly chose otherwise.

Ganapathy did not penalise us. He pushed us to defend it.

That, in itself, is rare.

We approached more than ten contacts. We got four real conversations.

A senior advocate in Madurai. A criminal law advocate at the Supreme Court.

The first was barely useful. He told me to look up cases in Tamil Nadu and walked off. The second gave us the data points that reshaped the problem: recruitment quality, geographic patterns, the Kerala anomaly, the D.K. Basu Guidelines that exist on paper and are unenforced in practice.

The field calls were telling us something the slides could not. Our customer was not the police department. It was a specific kind of officer inside the department. Reform-oriented. Blocked. Looking for a partner.

Great ideas die in the gap between value and adoption. We mapped the journey for our entry persona. Eight stages. The exercise produced one sentence we kept returning to.

“Anand learns about ARAM at the Police Club, the high-trust daily meeting point for officers. He realises ARAM is not just a civilian NGO. It includes retired police officers, signaling they speak the language of the force.”

That single sentence reorganised our partnership strategy. A retired DySP on the founding team was no longer a nice-to-have. It was the resource without which nothing else worked.

The impact-versus-paying-customer distinction. For NGOs, the question is not who pays for value received. It is who funds the work because they want the work to exist.

We built the donor matrix. CSR for ESG reporting. International agencies because India is on the Global Torture Index 2025. HNIs and retired judges acting on personal conviction. Crowdfunding for the public outrage moments.

Then the naming. We did not pick names for branding. We picked them because they told us what the work was for.

ARAM is the Tamil word for righteousness. The ethical foundation of a good life.

Kural means ethical voice in Thirukkural. The anonymous helpline for both victims and constables.

Oli means echo in Tamil. The audio monitoring system in custody rooms. What happens inside cannot stay hidden.

Ten lakhs installs accountability in five pilot stations and builds the first evidence pipeline on custodial violence in India.

While the donor matrix happened, the technology was being built.

A ten-stage audio pipeline. Voice activity detection. Voice emotion using MFCCs and LPC Residual Phase from Indian research on Indian speech. Whisper transcription on CPU. Trigger words that bypass the ML for help and get off. Claude Haiku as the fusion layer. A distress score zero to one hundred. A behavioural anomaly check against each speaker's own baseline. Live broadcast to a dashboard.

Score above seventy flips the room status to triggered.

The acoustic boost uses something called LPC Residual Phase, from Dr. N. J. Nalini's research on speech emotion recognition. Standard MFCCs capture the vocal tract. Residual Phase captures the raw glottal signal before filtering. Under emotional stress, the excitation source becomes more chaotic. That chaos is mathematically detectable.

The fusion rule is the line I keep returning to. Always escalate, never downgrade.

If the text says fearful but the voice says neutral, the system takes fearful. False positives are acceptable. Missed distress is not.

This is a design decision, not a technical one. It comes from the same place ARAM does.

Ganapathy opened with one sentence. Completion of activities is not a result.

The four levels.

Outputs are activities done.

Use of outputs is what people do with what you delivered.

Outcomes are behaviour changes in stakeholders.

Impact is system-level change.

The one indicator that proves it worked, in year three: a constable from a station ARAM never entered asks for Kural because a colleague told them it protected them.

Demand from inside the system. That is the proof.

Across seven sessions the problem was redefined seven times.

The training framing died in session three. The customer assumption died in session four. The civilian-NGO assumption died in session five. The branding-first assumption died in session six. The activity-equals-result assumption died in session seven.

Falling in love with the problem is not a slogan. It is a discipline of letting the problem reshape you faster than you reshape the problem.

Five students with no funding designed an NGO and built a working AI system for it in eight days. The barrier was never the technology. The barrier was the framing.

Development management gives you the question worth answering. AI gives you a way to answer it that did not exist five years ago. Together you get an evidence pipeline that did not exist in the country before.

That is the news.

Real teams are messy.

People got sick. Madhi had a headache. Shruti had a fever and a Lighthouse interview. Charlie was on a flight without electricity. Udaya was fixing a broken rack at home, then cooking for Justin who had food poisoning. One of us missed a meeting we had agreed to and Shruti said it plainly. I felt it quite. Udaya apologised. We moved on.

We did not call the right lawyer the first time. The first one told me to look up cases in Tamil Nadu and mentioned broken legs and walked off. We approached five contacts to get one good conversation. We rewrote the problem statement three times in 48 hours. co-ordinator asked for the submission again and again.

Every assignment got submitted. Every deadline got hit.

Not because we were impressive. Because whoever could carry it that day, did.

Sessions 8, 9, 10 are ahead. Financial model. Unit economics. Final pitch on April 24.

If you sit on a CSR committee, work with international human rights funders, know retired judiciary, work in police reform, or know a retired DySP who would talk to five students, write back.

We are listening. The system is listening too.

That is the whole point.

ARAM (Rarest of Rare) | Group | Social Entrepreneurship Elective | ISDM Noida | April 2026 Nami, Udaya, Charlie, Madhi, Shruti Faculty: P. R. Ganapathy.