Enterprise AI is moving fast. Large language models are being wired into decision pipelines, analytical workflows, and customer-facing systems across every major industry. But as Tejas Pravinbhai Patel sees it, the industry has been so focused on what AI can do that it has largely overlooked a more consequential question: who is responsible when it goes wrong.
Mr. Tejas Pravinbhai is a Senior Software Development Engineer at Amazon, where his engineering work touches the daily experiences of millions of customers across retail and AI platforms at global scale. He is an IEEE Senior Member a distinction conferred through a competitive elevation review by a committee of peers recognizing sustained professional achievement and Chair of the Irving ACM Chapter. He has presented research at IEEE conferences across North America, Europe, and Asia, and chaired sessions at international venues from Bahrain to Tunisia to Thailand.
Last December, at the International Conference on Computer and Applications (IEEE/ICCA 2025) hosted at Arab Open University in the Kingdom of Bahrain, that body of work was recognized with the conference’s Best Paper Award selected independently by three expert reviewers from a globally competitive field. The winning paper: Model Context Protocol Business Analyst (MCP-BA): A Governed and Explainable Framework for Enterprise Analytics.
For Mr. Tejas Pravinbhai, the recognition validated something he has long argued from his position inside one of the world’s largest technology companies.
The question was never whether AI is capable enough. The question is whether enterprises can trust it enough to actually rely on it and trust requires governance built in from the start, not bolted on at the end.
The Problem Tejas Pravinbhai Patel Set Out to Solve
Mr. Tejas Pravinbhai is direct about what is broken in enterprise AI today. Across his years of engineering experience at scale, he has observed the same pattern repeat: organizations deploy capable AI systems, then discover too late that those systems operated without identity enforcement, without audit trails, and without any mechanism to explain why a particular insight was generated or whether it was based on authorized data.
He identifies three structural failures that hold enterprise AI back from the high-trust deployments that regulated industries actually need. The first is a lack of KPI consistency different business units operate on conflicting definitions of the same metrics, and AI systems have no way to resolve or even detect the contradiction. The second is ungoverned data access models invoke tools and query databases without policy-controlled execution or identity binding. The third, and in Mr. Tejas Pravinbhai’s view the most damaging, is the complete absence of audit traceability.
Mr. Tejas Pravinbhai puts it plainly:
If an AI system produces an insight and you cannot reconstruct exactly what data it accessed, under whose authorization, and through what reasoning path you do not have an enterprise AI system. You have a liability.
That observation, grounded in the realities of production engineering at Amazon scale, is what motivated the architecture at the center of his award-winning research.
How Tejas Pravinbhai Patel Built the Answer: The MCP-BA Framework
Mr. Tejas Pravinbhai’s solution is the Model Context Protocol Business Analyst MCP-BA a governed reasoning framework that operationalizes the Model Context Protocol to connect user intent, organizational data, and analytical tools under identity-aware, policy-controlled execution. The framework was co-developed with a team of researchers spanning Oracle, Salesforce, Tavant Technologies, and the University of Illinois Urbana-Champaign.
The architecture Mr. Tejas Pravinbhai and his team designed has five coordinated layers. Every analyst interaction begins with an authenticated session bound to an organizational identity token governance starts at the first keystroke, not as an afterthought. A context broker layer then enforces role-based access control before any query reaches the data. Below that, Mr. Tejas Pravinbhai’s team built what he considers the framework’s most important innovation: a crawler–vector layer that continuously indexes governed enterprise assets dashboards, schemas, business glossaries into a live semantic context store, ensuring the AI always reasons from current, authorized information rather than stale embeddings.
Mr. Tejas Pravinbhai describes the design philosophy behind it: “Most systems treat the vector database as a passive repository. We treated it as an active governance substrate. Every piece of context the model touches has a lineage, a timestamp, and an access policy attached to it.”
Every query the system generates is executed through authorized MCP tool calls and bound to verified KPI definitions. And crucially, every insight produced leaves a complete, immutable audit trace reconstructable after the fact by compliance teams, auditors, or regulators.
The Results That Earned Best Paper Recognition
Mr. Tejas Pravinbhai’s team validated MCP-BA against two industry baselines a standard retrieval-augmented generation pipeline and Apache Atlas in a synthetic enterprise environment modeling a mid-sized retail organization with 120,000 sales records, 15,000 inventory SKUs, and a 1.2 GB corpus of internal governance documentation.
The performance gap was substantial across every dimension that matters to enterprise deployment. Context precision improved by 17 percentage points. Governance accuracy the share of tool calls executed within valid policy scope improved by 42 percent. Audit completeness reached 100 percent. And despite the additional governance overhead, reasoning latency fell by 29 percent, because policy-controlled execution eliminated the wasted cycles of unconstrained query retries.
Mr. Tejas Pravinbhai notes that the latency result surprised even his team: “People assume governance adds friction. What we found is that when you enforce context boundaries properly, the model stops hallucinating its way through unauthorized data sources and actually becomes faster. Governance and performance are not in tension they reinforce each other.”
A business impact simulation projected $185,000 in annual cost avoidance for a 50-analyst department, driven by a 38 percent reduction in manual reconciliation effort and a 60 percent decline in ungoverned data queries with no model retraining required.
Why Tejas Pravinbhai Patel Believes This Changes the Enterprise AI Conversation
For Mr. Tejas Pravinbhai, the significance of MCP-BA extends well beyond the benchmark results. He argues that the research demonstrates something the industry has been reluctant to accept: that governance, explainability, and compliance are not constraints imposed on AI systems from the outside they are performance features when engineered correctly from the inside.
Mr. Tejas Pravinbhai is emphatic on this point: “The EU AI Act, the NIST AI Risk Management Framework, U.S. executive orders on trustworthy AI these are not going away. Every enterprise deploying AI in a regulated context is going to face these requirements. The organizations that build governance into their architecture now will have a competitive advantage. The ones that treat it as a compliance checkbox will spend years retrofitting systems that were never designed for accountability.”
Mr. Tejas Pravinbhai also positions MCP-BA as complementary to, rather than competitive with, existing frameworks like AutoGen and DSPy. Where those systems excel at task decomposition and pipeline optimization, MCP-BA provides the governed execution substrate they lack the layer that ensures every action taken by an agentic pipeline is identity-bound, policy-controlled, and auditable. In Mr. Tejas Pravinbhai’s architecture, intelligence and accountability operate together.
Tejas Pravinbhai Patel: Engineer, Researcher, and IEEE Leader
Mr. Tejas Pravinbhai’s research career reflects the same conviction that drives his engineering work: that the most durable advances in AI come from systems designed for real-world accountability, not just benchmark performance. He has authored and co-authored papers across LLM inference optimization, GPU memory scheduling, distributed systems, and agentic AI architectures work that has been accepted at IEEE conferences and published in SCOPUS-indexed proceedings.
His standing in the international research community is recognized at multiple levels. As an IEEE Senior Member, Mr. Tejas Pravinbhai has cleared an elevation process that fewer than ten percent of IEEE members achieve one that requires demonstrated sustained contributions to the profession evaluated by a committee of senior peers. He chairs the Irving ACM Chapter, organizes IEEE sessions, and has served as session chair at international conferences across three continents in the past year alone.
At Amazon, Mr. Tejas Pravinbhai works at the intersection of his two worlds building production systems that serve millions of customers while contributing original research to the academic community. It is that combination, he says, that makes the governance problem personal.
When you build systems at Amazon scale, you understand viscerally what it means for an AI to operate without guardrails. The stakes are not theoretical. Millions of customer decisions flow through these systems every day. That reality shapes how I think about research and it is exactly why governance has to be engineered in, not wished in.
About the IEEE/ICCA 2025 Conference
The International Conference on Computer and Applications now in its seventh edition under the IEEE Bahrain Section is a globally recognized venue for computing research, drawing submissions from across disciplines and continents. The 2025 event was held at Arab Open University in the Kingdom of Bahrain from December 22 to 24, chaired by Professor Dr. Jihad Mohamad ALJAAM. The Best Paper designation is awarded through independent evaluation by three domain experts drawn from the conference’s international review pool making the recognition a genuine peer judgment rather than a ceremonial one. Mr. Tejas Pravinbhai’s paper was among a small handful selected from the conference’s full accepted proceedings.
Mr. Tejas Pravinbhai also served as Session Chair at the conference contributing to the event not only as an awarded researcher but as a trusted member of the international IEEE community responsible for guiding the scientific program.
.
:::tip
This story was distributed as a release by Jon Stojan under HackerNoon’s Business Blogging Program.
:::