From the Engineering Floor
Deep-dives into sovereign AI, RAG architectures, agentic workflows, and what it actually takes to ship enterprise AI in production. Written by the engineers and leaders building it.
The 2026 Sovereign AI Architecture Report
Data-driven analysis of every meaningful sovereign AI stack in production today. Compares 6 open-weights model families, 4 vector …
State of Agentic AI in Regulated Industries 2026
A production-pattern survey of agentic AI in BFSI, healthcare, public sector and pharma. What patterns actually ship (ReAct + tool…
EU AI Act Readiness Benchmark — 50 Enterprises
Anonymised readiness benchmark across 50 enterprises with EU exposure — banks, insurers, hospitals, manufacturers, public-sector b…
What CRO Conversations on AI Look Like in 2026
Synthesis of 50+ Chief Risk Officer conversations across BFSI, healthcare and public sector over the past nine months. What they'r…
The Sovereign AI Inflection Point: Why Regulated Enterprises Are Moving On-Prem
Central banks, insurers and healthcare systems now insist their AI models run on their own infrastructure. The driver isn't fear o…
RAG on Your Own Servers: Architecture Patterns for Air-Gapped Enterprises
Building a RAG system inside a regulated bank or hospital is a different sport. The cloud tutorials don't translate, and the failu…
NASSCOM Tech Excellence 2026: How We Built the Healthcare AI Stack
Our NASSCOM Tech Excellence 2026 win recognised the Healthcare AI Stack we shipped over the last four years: Rx Compliance Stocker…
MindMap Acquires Bluetide.co: Deepening Our Data and Agentic AI Stack
We're excited to announce that MindMap Digital has acquired Bluetide.co, a specialist data engineering and agentic AI firm. Here's…
NL-to-Dashboard: 6 Patterns That Work (and 4 That Don't)
Natural language to dashboard sounds simple. In practice, it breaks in six distinct ways. After 30+ BI deployments, here are the p…
From RPA to Agentic AI: The Operating Model Shift Nobody Talks About
RPA solved repeatable rule-based tasks. Agentic AI solves judgment-based tasks. The shift isn't technical — it's an operating mode…
KYC at Scale: The OnboardX Architecture That Onboards 50,000 Customers Monthly
OnboardX processes 50,000+ banking KYC applications monthly across Africa and the Middle East. Here's the technical architecture —…
AI Voice Agents Are Replacing Call Centre Teams. Here's How.
We've deployed AI voice agents that handle outbound collections, inbound support, and appointment booking at costs 60% below human…
Introducing Agentic Workflow Studio: Build Multi-Agent Enterprise Workflows Without Code
We're launching Agentic Workflow Studio — a drag-and-drop builder for multi-agent enterprise workflows. Design, test, and deploy a…
What the EU AI Act Means for Credit Scoring
Credit-scoring AI is explicitly Annex III high-risk under the EU AI Act. From 2 August 2026, every bank or lender whose scoring de…
Sovereign AI vs Bring Your Own Key: Why BYOK Doesn't Close the Compliance Gap
Bring Your Own Key gives the customer cryptographic control over data at rest. It doesn't give them control over the model, the in…
RAG vs Fine-Tuning for Enterprise: A 2026 Decision Framework
Two years into the enterprise LLM era, the RAG-or-fine-tune debate finally has a defensible answer: do retrieval first, exhaust it…
What It Actually Costs to Run Llama 3.3 70B On-Prem
Cloud LLM pricing has dropped twice in 2026. On-prem economics dropped further. Below ~200M tokens per month the cloud is still ch…
Clinical AI Is High-Risk. Now What?
Clinical decision support systems are Annex III high-risk under the EU AI Act. By 2 August 2026, every hospital, health system and…
The Article 25 Trap: When Using AI Quietly Makes You the Provider
Most enterprises think they're deployers of AI, not providers — and therefore subject to a lighter set of EU AI Act obligations. A…
ReAct in Production: When the Loop Earns Its Cost
The ReAct pattern — alternating Thought/Action/Observation triples — is the foundational agent loop. It's also the most over-appli…
What 94% Straight-Through Processing Actually Requires
Every IDP vendor's demo promises 95%+ straight-through. Every production IDP deployment that's been running for more than 9 months…
Vector Database Comparison: pgvector, Qdrant, Milvus in 2026
The vector database decision is now operational, not architectural. The three open-source options — pgvector, Qdrant, Milvus — all…
LLM Observability with Langfuse: What to Capture, What to Ignore
LLM observability has matured from "log everything to a JSON file" to a proper discipline in 2026. The substrate for sovereign dep…
AI Compliance for Tier-1 Banks: RBI, SAMA, EU AI Act in One Picture
A tier-1 bank serving customers across India, the Gulf and the EU now has three overlapping AI-regulation regimes converging on th…
How to Evaluate a RAG System: The Four Numbers That Actually Matter
RAG evaluation is the single highest-leverage engineering discipline in enterprise LLM work, and the one most teams skip. The four…
Prompt Injection — The SQL Injection of LLMs
Prompt injection isn't a hypothetical — it's the #1 vulnerability in OWASP's Top 10 for LLM Applications, and the production attac…
Building Human Oversight into Agentic AI (What Article 14 Actually Means)
Article 14 of the EU AI Act requires effective human oversight of high-risk AI systems. The translation from regulatory language t…
Why HIPAA Timelines Are Quietly Pushing Healthcare AI On-Prem
HIPAA permits cloud LLM use under a Business Associate Agreement, in principle. In practice, the BAA review process at US covered …
Schema-Driven Extraction Beats Template OCR — When and Why
Template-based OCR with field coordinates works on the 60% of enterprise documents that are highly structured. It breaks on the 40…
Sovereign Chatbots for Citizen Services: What 12-Language, Air-Gapped Production Looks Like
We deployed a citizen-service AI assistant for a national government agency last year — fully air-gapped, twelve languages, every …
Choosing Between Llama 3.3, Qwen 2.5 and Mistral for Enterprise: The 2026 Cut
The three open-weights families that matter for enterprise deployment in 2026 are Llama 3.3, Qwen 2.5, and Mistral's recent releas…
The 90-Day Path to EU AI Act Compliance — A Programme Guide for the CIO
Sixty days remain before the 2 August 2026 deadline. Most enterprises are behind. Most won't be ready. A defensible posture is sti…
AI for KYC: What 5 Days to 4 Hours Actually Looks Like in Production
OnboardX collapsed an East African tier-1 bank's KYC cycle from a 5-day median to under 4 hours, with dropout falling from 42% to …
Building an AI Centre of Excellence That Actually Ships
Most enterprise AI Centres of Excellence we've reviewed produce strategy decks. The ones that actually ship production AI follow a…
Open-Weights vs Closed API: The Architecture Decision That Defines 2026
Two years ago the open-weights vs closed-API decision was a capability discussion. In 2026 the capability gap on enterprise worklo…
GPAI Obligations Explained: What Articles 53 and 55 Actually Require
The EU AI Act's GPAI provisions — Article 53 (baseline obligations on all general-purpose AI model providers) and Article 55 (addi…
The Five BFSI AI Deployments That Defined 2025 — And What They Teach for 2026
Five deployments at tier-1 banks across three continents represented, by our reckoning, the most ambitious BFSI AI work of 2025. E…
Want to work with us?
117 accelerators. First pilot in 6–9 weeks. Sovereign by design.