How the work is done

Most of my work begins with a decision someone has to take. A police leadership team has to know where a protection cell is leaking. A donor has to know whether to buy help desks or dealer incentives. A monitoring cell has to know whether an enrolment count is standing in for agency. The method follows that decision.

Qualitative research, for me, is semi-structured interviewing and sitting with a process until it becomes visible. At Pride Place that meant case-record review and watching intake. At Sanchay it meant following a pensioner toward a BDO office. For TRI it meant fourteen interviews with SHG members, panchayat representatives and practitioners. I write interview guides. I try to code themes rather than harvest quotations. When the codebook is not public, I say so.

Institutional and legal analysis is the other half of that fieldwork. A statute is a set of duties, not a mood. The Transgender Persons Act 2019, the NFSA, MGNREGA guidelines, pension circulars: I read them as instructions that some office will have to perform. Implementation research is the comparison between that instruction and the ordinary day.

Quantitative work, on the public pages, is at the scale of published tables and coursework models. I use NFHS factsheets and national reports, and Stata for ordinary least squares on those tables. I can run a difference-in-differences in a classroom. A portfolio page gets one only when the design supports it. Excel is the stronger daily instrument: lookups, pivots, scenario and sensitivity analysis, programme cost and coverage models. The migrant costing workbook is the public example.

Spatial work, at this stage, is state-level mapping of published indicators. The phone-gap cartogram is that practice: a policy question first, a map that answers it, a CSV that could be dropped into QGIS. Python and SQL are beginner skills on the résumé. The workflow page treats them that way: cleaning, joining, aggregating, writing out a file someone else could check.

Strategy is the habit of refusing a recommendation until the diagnostic has earned it. Problem, decision-maker, hypotheses, evidence, options, criteria, operating model, measurement, risks. I have learned to distrust verbs such as “strengthen” unless they come with a who, a where, a when and a number. Communication is part of the method. A deck for police leadership and an essay for Village Square are different objects. Both have to be true.

I use Cursor and Claude as research assistants: to organise sources, to draft tables, to build small tools such as the evidence matrix. I do not outsource the judgement about what a number means. If AI touched a page, the artefacts still have to stand without it.