Bengaluru · Pre-seed & Seed AI Research

I study the founders turning AI capability into products customers rely on.

Founder/operator based in Bengaluru, researching memory and context, reliable AI deployment and focused business workflows.

01 / Selected judgment

Three companies. Three tests.

Start with the conclusion, then inspect the evidence and the condition that would change the view.

Pre-seed · EuropeInvestigate

Qontext

Does permission-safe shared context improve task success beyond ordinary retrieval when facts and access rights change?

Evidence
Public quickstart documentation describes ingestion, connectors and agent access through MCP or API, making the proposed product boundary inspectable.
Open question
Whether shared context beats a well-maintained retrieval baseline after changes, revocations and contradictory updates, at a defensible total cost.
Read the screen ↗

Assessment · Reviewed 2026-11-04
Public research · Memory & context

Pre-seed · IndiaInvestigate

SuperBryn

Do evaluations predict human-adjudicated failures across accents, noise and code-switching, and which capabilities are shipped?

Evidence
Kalaari's original investment announcement connects the founders' operating and speech-research backgrounds to a specific testing and observability product.
Open question
Whether automated judgments catch material real-world failures that customers would otherwise miss, without creating an unaffordable review queue.
Read the screen ↗

Assessment · Reviewed 2026-11-04
Public research · Reliable deployment

Seed · IndiaInvestigate

Neurofin

What material fraud or compliance defects are missed, how are exceptions audited, and how much human review remains?

Evidence
The current site describes a defined document-checking workflow and publishes named customer references, creating concrete claims that can be checked with permission.
Open question
Whether reduced turnaround and rework persist after accounting for missed material defects, manual exceptions and customer-specific implementation.
Read the screen ↗

Assessment · Reviewed 2026-11-04
Public research · Business workflows

Conditional research assessments, dated to the evidence reviewed.

02 / Where I look for an edge

Scarcity moves as intelligence becomes abundant.

01

When is context worth a separate budget?

Look for permission-safe memory that improves a real task beyond simpler retrieval.

Explore memory & context →
02

Can reliability become a product?

Look for tests that predict production failures and reduce the cost of a successful task.

Explore reliable deployment →

03 / Regional research

Follow the evidence across markets.

India is my relationship base. The UAE, Europe and USA are research coverage regions.

QontextEurope
Pre-seed
Announced 2026-02-05
Proposed diligence: compare native context, simple retrieval and Qontext on held-out tasks with changing facts and access rights.
SuperBrynIndia
Pre-seed
Announced 2025-12
Proposed diligence: compare scores with blinded human labels on held-out calls; verify shipped features and review time saved.
NeurofinIndia
Seed
Announced 2025-07-10
Proposed diligence: use a blinded labelled document set; measure material misses, exception auditability and human minutes per file.

11 unique records in the published research pool. Counts exclude verification holds. Region membership can overlap. Explore the database →

Coverage gap: no standalone memory company cleared both geography and stage evidence for India or UAE in the supplied research.

04 / From building to diligence

Operator evidence / Perfstaq

Where should the human approval sit?

Perfstaq connects captured founder context with business workflows. Its public product surfaces make approvals part of taking an action.

Inspect the operator case →

Build the workflow. Question the outcome.

Working across AI product development, customer discovery and commercial iteration gives me a practical lens on integration effort, failure recovery and what the user still has to do.

I attend startup events in Bengaluru and connect with founders through LinkedIn. Published company research here uses public sources.

Open the product artifact ↗

05 / Recent notes

Arguments you can inspect.

Research notes connect the product question to a testable business question.

Memory & context · 4 min read

When does memory become a standalone business?

Memory earns an independent budget when it keeps changing knowledge usable across applications and measurably improves work beyond simpler alternatives. Persistence alone is an insufficient investment thesis.

Read article ↗

Published 2026-11-04

Reliable deployment · 4 min read

The cost of a successfully completed AI task.

The useful unit is the accepted business outcome. Retries, external tools, review and failed work belong in its cost, even when the model call looks cheap.

Read article ↗

Published 2026-11-04

Open the journal →

Start a conversation

Have a difficult early-stage AI question?

Let’s discuss the evidence, the unknowns and the next useful test.