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
Where does software own the work?
Look for repeated use, a clear buyer and manageable implementation and review costs.
Explore business workflows →03 / Regional research
Follow the evidence across markets.
India is my relationship base. The UAE, Europe and USA are research coverage regions.
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.
Pre-seed
Announced 2025-12
Proposed diligence: compare scores with blinded human labels on held-out calls; verify shipped features and review time saved.
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.
A contract changes→Context is corrected→Access is checked
Memory & context · 4 min readMemory 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
Define the assigned task→Accumulate every attempt→Record intervention
Reliable deployment · 4 min readThe 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 →