Skip to content

MEMORY — Running Decision Log

This file tracks decisions made across sessions. Any AI agent working in this repo should read this to understand what's been decided and what's still open. Update this file whenever a meaningful decision is made.


Decisions (chronological)

2026-05-16 — Foundation session (Session 1)

Context: First session establishing PropPie's AI foundation. Aishvarya (COO) driving; Vishal (CEO) building pipeline in parallel.

# Decision Rationale Reversible?
D1 Drop "first AI proptech in India" positioning. Replace with "India's first compliance-grade, source-cited, probabilistic real-estate intelligence layer." Factually wrong — Housing.com, NoBroker, PropEquity all use AI. The honest framing is defensible; the "first" framing is not. Yes (marketing copy)
D2 Info-only stance for B2C Broker. No personalised investment advice. No SEBI IA/RA registration. Fastest to market; aligns with Honest Broker values; avoids regulatory risk. Still enables comparison, simulation, contextualisation, flagging. Yes — can register as SEBI IA later if needed
D3 Maharashtra-first. No pan-India data or product expansion in v1. Depth > breadth. Maharashtra has the best public data infrastructure (MahaRERA, IGR). Moat is depth per micromarket. Yes — expand later
D4 Defamation guardrails: percentile framing, citation, right of reply. Never "bad developer" — always "78th percentile delay vs peers." Legal risk; also aligns with honesty stance. No — this is a permanent discipline
D5 Three products on one data foundation. PropPie Fractional (live) + Broker (B2C) + Analytix (B2B) all share the canonical attribute store. Efficiency for a 5-dev team; derived attributes compound across products; ground truth flows back from Fractional. Partially — products can diverge, but shared data layer is load-bearing
D6 Vishal owns pipeline architecture. This repo defines the contract (what to produce), not the how. Clear ownership; avoid design-by-committee on infra. Yes
D7 ~140 attributes refined from original 92. Added categories: Investor Persona, Macro Overlay, Fractional-specific, expanded Risk Intelligence, Tenant & Lease. Added operational columns (confidence, lineage, freshness). Gaps in original spec; operational columns needed for quality and Honest Broker citations. Yes — attributes can be added/removed
D8 Source-of-truth hierarchy: IGR > MahaRERA > MahaBhulekh > GRs > GIS > Internal > Licensed > News > Social > Listings. Registered deeds are the most authoritative source for price; RERA for project metadata; etc. Adjustable with evidence

2026-05-17 — Foundation session continued (Session 2)

Continued document creation. All product specs, AI architecture, and memory layer established.


Current state

Item Status Last updated
AGENTS.md + README.md Complete 2026-05-16
5 Cursor skills Complete 2026-05-16
SOUL + tone-and-voice Complete 2026-05-16
Strategy docs (5 files) Complete 2026-05-16
Data docs (6 files) Complete 2026-05-17
Product docs (5 files) Complete 2026-05-17
AI architecture (3 files) Complete 2026-05-17
Memory docs (4 files) Complete 2026-05-17
Extraction schema v2 Exists (Vishal) Pre-session
Pipeline implementation In progress (Vishal) Active
Application code Not started

Conventions

  • Append new decisions with date and session number
  • Use D# numbering (D1, D2, ...) for easy reference
  • If a decision is reversed, don't delete — strikethrough and add the new decision with rationale
  • Include "Reversible?" — helps future sessions assess flexibility