Not passive.
Not quite aggressive.
Just research-first.

I studied economics. The thing that stuck with me from those years wasn't a particular model. It was a conviction that firms are not interchangeable. A small firm in its first decade faces a fundamentally different problem set than a mature large-cap. The factors that drive success at one stage of a company's life cycle are not the same factors that drive success at another. Theory of the firm, industrial organisation, network effects all tell us that the right lens depends on what kind of firm you are looking at.

Most retail investors in India are told to choose between two extremes. Passive index funds on one side, speculation on tips on the other. Neither approach uses what economics actually tells us about how firms create value. PAI is the framework I built for the middle path. Research-driven, framework-led, transparent.

I draw on what I learned as an economics student, and on time spent working in lending. Particularly in MSME lending, where you see up close how cash flows, customer concentration, working capital cycles, and informal networks actually decide whether a small business survives. Those years made me sceptical of valuation models that ignore the network around a firm. A balance sheet tells you what a company owns. It does not tell you who its customers are, who its investors are, or whether the founder has the relationships to keep the lights on through a downturn.

That observation, that network effects are systematically underweighted in standard valuation and technical frameworks, became the foundation of the C2 layer in PAI. The thinking on this layer was sharpened in conversations with a friend pursuing a PhD in economics whose research focus is on networks. I am grateful for those conversations. The framing of network quality as a scorable layer rather than soft commentary owes a lot to them.

Most investment writing online does not show its own positions. Reports go out, calls get made, and the author is invisible when those calls go wrong. I find that uncomfortable. If I am putting research out under my own framework, I should also show what that framework looks like when I act on it myself.

My portfolio is on this site for that reason. The wins are visible. The losses are visible too. EIEL is up 31% from my entry. Reliance is down 55% from where I bought it. Aelea Commodities is in the avoid bucket and I still own it because I bought it before I had the framework. All of that stays on the page.

There is a second reason for showing it. Every model has limitations. No framework gets every call right. The score is a directional signal, not a guarantee. A high PAI Score raises the probability of a good outcome. It does not promise one. Putting my own portfolio next to the model output is the most honest way to show that gap. You can see when the model worked and when it did not. You can see when I followed the framework and when I did not.

And there is a third reason that often gets overlooked. Execution matters as much as the call. Buying at the right level, holding through volatility without panic-selling, exiting when the thesis breaks rather than when the price wobbles, all of that is separate from whether the underlying research is sound. The same buy-zone call, executed well or executed badly, produces very different returns. My portfolio is a record of execution, not just of analysis. It is one thing to identify a Stage 2 setup. It is another to actually buy it on the day, hold it through a 10% drawdown, and not flinch.

I am not selling anything. I am not running a paid newsletter, a Telegram channel, or a tip service. PAI exists because I needed it to exist for myself. The site is here so other people can read along if they find the work useful. The portfolio is here so they can see what acting on the work actually looks like.

The π in PAI is not decoration. This work would not be possible at this depth, at this speed, by one person, without AI. I want to be transparent about how AI is actually used here, because the difference between "AI-powered" as marketing and AI as a real working tool is large.

Specialised AI agents handle different parts of the pipeline. Each agent has one job, and the framework binds the outputs together into a coherent thesis.

• 0
  News and idea scout. Continuously scans financial news, sector announcements, government scheme updates, and market chatter. Surfaces sectors and companies that look worth a closer look. This is the funnel that feeds everything else.
• 1
  Sector and policy analyst. Reads government schemes, ministry circulars, and budget documents. Builds the L1 regulatory case that anchors the NAV framework.
• 2
  Universe builder and network mapper. Pulls financials, identifies peer sets, maps customer concentration, institutional holdings, and promoter networks. This is where the C2 layer comes from.
• 3
  Quadrant scorer. Applies the framework rules to assign each company to a quadrant and compute the PAI Score components.
• 4
  Peer comparator. Benchmarks each candidate against its sector peers on every relevant metric, surfaces who is genuinely better and who is just cheaper.
• 5
  Deep dive writer. Pulls everything together into a structured report. Same template every time, no narrative drift.
• 6
  Technical analyst. Computes moving averages, evaluates the trend template, classifies the stage, identifies pivots. Replaces hand-drawn chart analysis with consistent, rule-based output.
• 7
  Live verification monitor. Checks every price-sensitive number against a live source before publication. Catches stale data before it becomes a stale call.

Every output goes through me. I read the report, push back on the score where I disagree, override the agent when I think it has missed context, and decide what gets published. The AI is fast and consistent. The human supplies judgment, scepticism, and the willingness to throw an output away when it does not feel right.

This is also why the framework can be honest about its limits. When AI is doing the heavy lifting on consistency, the human can spend energy on the things AI is bad at. Asking whether the data is even the right data. Noticing when a company's narrative has changed in a way the numbers have not yet caught. Deciding when a high score should still not be acted on because something feels off.

If different kinds of firms need different lenses, no single framework can be universal. NAV is the first one. It is designed for small and mid-cap firms riding identifiable government policy tailwinds, where the regulatory anchor is a meaningful driver of revenue. It would be poorly suited to a large-cap turnaround, an early-stage IPO, or a momentum-driven breakout in a sector with no scheme behind it.

That is why this site is architected to be framework-modular. As I add new frameworks for special situations, momentum, large-cap re-rating, early-stage IPOs, they will appear as separate scoring systems. Each will have its own conviction bands, its own report format, and its own published track record. Reports will be tagged with which framework was used. Conflicting calls between frameworks will be visible, not hidden.

The PAI Score itself is proprietary. The exact rubric for converting layer assessments into a single number is not public. What is public is the score, the band it falls in, and the reasoning behind each layer for each company.

The current focus is finishing coverage on sectors where NAV applies cleanly. Water, cables, recycling, auto metal forming, lending, energy. Reports will be added continuously. Beyond that, additional frameworks for different firm types are in design. Each will get its own scoring system, its own published track record, and its own home on this site.

I am also working on backtesting methodologies for the framework. The most immediate test is rebuilding the NAV scores using only data available as of 31 December 2024, then measuring how those calls would have performed across calendar 2025. That gives a full year of out-of-sample returns to look at, with no hindsight bias. Done honestly, this kind of test will show where the framework actually works and where it overfits. The results will go on this site, whatever they say.

Two new AI agents are also in development, extending the existing pipeline beyond entry-side analysis into ongoing portfolio monitoring. A good entry call only matters if the exit is also handled well.

• A1
  Adverse news monitor. Continuously scans publicly available news and disclosures for every company in the active portfolio, the buy zone list, and the 75-plus score band. Flags governance changes, auditor exits, regulatory actions, and material announcements that should re-trigger a thesis review.
• A2
  FII/DII flow monitor. Tracks foreign institutional and domestic institutional flows at the sector and stock level. When institutional positioning shifts materially in a stock you own, the framework needs to know quickly. Knowing when to exit is as critical as knowing when to enter.

If a report on this site helps you think more clearly about a company, whether you end up agreeing with the call or disagreeing with it, that is the entire point. The framework is a thinking tool, not a buy list.