AI Research Scientist & Consultant at AFRY · US Patent Holder · International Speaker
I architect the transition from AI experimentation to production-scale impact — designing intelligent systems where AI becomes dependable operational infrastructure, not experimental software.
Based in Sweden. Speaking across 20+ countries. Building in the open.
FedShield — Hierarchical Federated Security Intelligence
Open-source implementation of US Patent 10,805,353 B2 — a three-tier hierarchical federated ML security system that protects enterprise networks through intelligent, resource-proportional model updates. All 20 patent claims implemented. Data never leaves the device.
| Feature | Detail |
|---|---|
| Patent Claims | 20 / 20 implemented |
| Architecture | 3-tier: Device → Cluster → Global |
| Data Privacy | Zero data shared off-device |
| Origin | Bank of America · Filed 2018 · Granted 2020 |
Athenium — Contract Risk Intelligence
A transformer that reads financial contracts the way an expert does. Built from first principles — attention implemented from scratch, LoRA fine-tuning on Mistral-7B, self-hosted inference. Every layer of the stack, explained.
| Metric | Result |
|---|---|
| Macro F1 | 0.971 |
| P95 Latency | 118ms |
| Auto-processed | 88.3% |
| Cost / 1K docs | $0.18 vs $14.20 (GPT-4o) |
At AFRY — one of Europe's largest engineering and consulting firms — I lead the AI Global Competence Group and work across industrial and enterprise environments, bridging advanced AI capabilities with real operational deployment.
Previously, I built financial data products and distributed ML systems at Bank of America's Chief Data Scientist Organization (GTO), where I co-invented a US Patent for a hierarchical machine learning framework for adaptive enterprise threat detection.
Core areas: Intelligent Systems Architecture · AI Governance & Safety · Production ML · AI Product Strategy · Strategic Advisory
I believe the best way to advance AI is to build in the open. These tools are born from real production experience, shared freely.
| Repository | What it does |
|---|---|
| Athenium | Contract risk intelligence — Mistral-7B fine-tuned with QLoRA, built from first principles. 0.971 Macro F1. Self-hosted. |
| FedShield | Open-source implementation of US Patent 10,805,353 B2. Three-tier hierarchical federated ML for enterprise security. Zero data shared off-device. |
| LLM-Semantic-Cache | Stop paying for the same LLM response twice. Cache by meaning, not exact text. |
| AI-Agent-Autopsy | Cut open broken LLM agents. Find what killed them. Fix it before you ship. |
| RAG-Autopsy | Cut open broken RAG pipelines. Find what killed them. Fix it before you ship. |
I write about production AI, future cities, governance, and what happens when intelligence stops living only inside humans.
Latest articles → prajitdatta.com/thoughtsandresearch
- The City That Debugs Itself: Inside the Real-Time Operating Systems Running Tomorrow's Megacities
- The Invisible Patient: How Continuous Biosensors Will Make "Feeling Fine" a Medical Diagnosis
- Why 94% of Companies Are Still Failing at Production Deployment
- When Silicon Valley Met Swiss Alps — and Neither Could Agree on AI's Future
📬 Subscribe for new articles → prajitdatta.substack.com
I deliver keynotes and workshops at technology and leadership forums worldwide — TEDx, CIO/CTO conferences, Almedalsveckan, CES, and executive summits across the US, Europe, and Asia.
Topics: Production AI · AI Governance & the EU AI Act · Intelligent Systems Architecture · The Future of Human-AI Collaboration
Booking a speaking engagement? Reach me at prajitdatta@outlook.com
Forbes Technology Council · World Economic Forum — Global Shapers · TEDx Speaker · Norden Impact — Board Member · Linux Foundation / Hyperledger — Sweden Lead · Google Developer Community — West Sweden Leader
| 🌐 Website | prajitdatta.com |
| 📬 Substack | prajitdatta.substack.com |
| in/prajitdatta | |
| ✍️ Medium | @prajitdatta |
| 📊 Kaggle | prajitdatta |
| 🤝 Work with me | topmate.io/prajitdatta |
| 🎤 Speaking enquiries | prajitdatta@outlook.com |
"Technology becomes valuable when humans can rely on its behaviour."
— Karmev Jayate