OldKnowledge,New Vessel.
Essays on on-device AI, data sovereignty, and building systems that keep knowledge where it belongs.
Enter the archive →— 00 / ThesisEvery business runs on knowledge older than its software.
Latest Entries
Generative AI Architecture Advisory: The Depth to Challenge, Not the Depth to Build
Consolidating Duplicate AI Capabilities: The Migration Nobody Plans and the Service Nobody Runs
Evaluation-as-a-Service: Stop Every Team Reinventing the Meaning of "Working"
Architecture Governance Across Multiple Delivery Partners: Reviewing Work You Don't Control
The AI Platform Architecture Blueprint: The Complete Enterprise Reference (2026)
अथ प्रथमोऽध्यायः ॥
विद्या ददाति विनयं विनयाद् याति पात्रताम् ।
पात्रत्वाद्धनमाप्नोति धनाद्धर्मं ततः सुखम् ॥
We digitize centuries-old manuscripts. Then we build with the same discipline.
AppScale's roots are in a quiet project: structuring classical Sanskrit texts into faithful digital form. Extraction, structure, provenance, sovereignty — the same principles now power our client work.
Built for Every Business
Intelligence that never leaves the device
Offline-first, data-sovereign systems — LLMs, vector search, inference at the edge.
02Document IntelligenceAir-gapped document vaults
Sealed, searchable, provable.
03AI ProductsFull-cycle AI product builds
Mobile-first, production-grade.
04MCP & AI SecuritySecuring the agentic stack
Tool poisoning, prompt injection, confused deputies — studied and hardened.
05Indic Language AIOCR for scripts the world forgot
Sovereign OCR for Indic scripts.
06Consulting & Architecture19+ years shipping at scale
OTT, e-commerce & mobility platforms serving millions.

Generative AI Architecture Advisory: The Depth to Challenge, Not the Depth to Build
GenAI advisory is depth to challenge, not build. The one failure each use case is prone to, the crux question per technical lever, and the discipline of holding the altitude.

Consolidating Duplicate AI Capabilities: The Migration Nobody Plans and the Service Nobody Runs
Detecting five duplicate AI capabilities is easy; collapsing them into one consumed service is not — pick the reference on evidence, migrate incrementally, run it as a product.

Evaluation-as-a-Service: Stop Every Team Reinventing the Meaning of "Working"
When every AI team builds its own evaluation, "working" means a dozen incompatible things. Make it a shared service: versioned datasets, governed judges, a CI regression gate.

Architecture Governance Across Multiple Delivery Partners: Reviewing Work You Don't Control
Governing AI architectures from multiple SIs and internal teams is a distinct discipline: review, challenge, approve and continuously conform work you influence but can't command.

The AI Platform Architecture Blueprint: The Complete Enterprise Reference (2026)
A complete 13-chapter enterprise reference for designing, building, and evolving an AI platform — the shared substrate that lets every team ship AI features without rebuilding the same plumbing. Covers the 7-layer platform model (data, model management, inference, orchestration, evaluation, governance, developer experience), 4 reference architectures (Startup, Growth, Enterprise, Regulated-Industry), the model lifecycle and registry, multi-tenant data isolation, semantic caching and routing, FinOps with attribution and budget guardrails, EU AI Act and HIPAA-aligned governance, and the Phase 0 to Phase 4 maturity roadmap with team sizes and timelines. Includes a 20-question platform readiness assessment and a 30-question FAQ optimized for the decisions enterprise architects actually face.

Enterprise AI Architecture & Strategy: The Authority Layer Every AI Program Is Missing
Enterprise AI fragments because every team optimises locally and nothing owns coherence. The fix is a senior architecture authority — its six responsibilities, mapped.

The Transformer Isn't Being Replaced. It's Being Hybridised
State space models like Mamba are linear-time alternatives to quadratic attention. But pure SSMs fail at exact recall, so production in 2026 means hybrids, not replacement.

Your Detections Are Software. You Just Never Tested Them Like Software
Detections written in a SIEM console are unreviewed, untested code shipping to prod. Detection-as-code fixes it with Sigma, CI tests, ATT&CK coverage and regression.

Zero-ETL Deletes the Pipeline You Operate, Not the Semantics You Inherit
Zero-ETL removes the pipeline you operate, not the eventual-consistency semantics, resync-on-DDL and single-region limits it inherits. Read the guarantees first.

The Request Was Authenticated. It Just Wasn’t Allowed to Read That Record.
BOLA is the #1 API vulnerability because it is a missing check, not a bug. The deny-by-default, object-boundary authorization architecture that kills it.

Cell-Based Architecture: Blast Radius Is a Number You Choose
Blast radius is chosen by architecture, not by the bug. How cell-based architecture, thin routers and shuffle sharding cap impact — and when not to cell.

The Payment Page Is Assembled in a Browser You Don’t Control
Your build pipeline is clean and your checkout still gets skimmed. Client-side supply chain security with strict CSP, SRI and PCI DSS 6.4.3/11.6.1.

CXL Memory Pooling: Reclaiming the DRAM You Already Paid to Strand
A quarter of your datacenter DRAM is stranded behind cores already sold. CXL memory pooling reclaims it — for the latency-tolerant workloads that actually fit.

The Image Was the Prompt: Multimodal Injection via Images, PDFs and Screenshots
Multimodal prompt injection hides instructions in images, PDFs and screenshots that text filters never parse. The 2025 CVEs, and the architecture that contains it.

Postgres as the Entire Stack: Queue, Cache, Search, Vector — and Where It Stops
Postgres can be your queue, cache, search engine and vector store. When consolidating onto one database is right, and the exit signal for each workload.

Your Coding Assistant Has Read the Whole Repo. Where Does That Context Go?
Your coding assistant sends the whole repo across the IDE-to-vendor boundary. Govern it with enterprise tiers, enforced retention, egress control and DLP.

Graph Neural Networks Are Quietly Winning Fraud and Logistics
Fraud rings and delivery networks are graphs. GNNs see multi-hop structure flat models miss — but full-graph inference on the hot path kills latency. Here is the architecture.

Agent Memory Poisoning: The Attack That Persists After the Prompt Ends
Prompt injection dies with the turn; memory injection persists. One poisoned record is retrieved for months. Authorise writes, provenance every record, quarantine.

Iceberg Won. Now the Fight Is Over the Catalog
Iceberg made the files portable. The catalog — commit pointer, RBAC, credential vending — is where lock-in moved. Polaris vs Unity vs Glue vs Gravitino, honestly.

Your Cloud Has 40,000 Effective Permissions. CIEM and the Entitlement Explosion
Identities use ~1% of the 40,000+ permissions clouds expose. CIEM computes effective permissions, baselines usage, and shrinks the gap — without breaking prod.
Bring this thinkingto your business.
conversation
One essay, most weeks. No noise.