I'm Ashwini Sharma — building enterprise knowledge layers for AI agents, defining North Star architectures for Generative AI, Agentic AI & RAG, and turning the art of the possible into production-grade platforms.
agent EnterpriseBrain {
knowledge: GraphRAG + VectorStore,
retrieval: HyDE → rerank → context,
reasoning: multi-agent orchestration,
guardrails: responsible-ai,
observability: LLMOps + evals,
}
// scaled beyond PoC → production
deploy EnterpriseBrain → cloud ✓Over 14 years across Information & Communication Technology — from processing billions of records a day for global telecoms to architecting the enterprise knowledge layer that powers autonomous AI agents today.
I advise C-level stakeholders on AI strategy and translate cutting-edge GenAI capabilities into organization-wide platforms that live beyond the pilot. My work spans the full arc: executive discovery → North Star architecture → production delivery, with a deep focus on governance, cost-efficient inference and LLMOps observability.
Alongside the engineering, I'm a published Explainable-AI researcher — my Master's thesis introduced RICE, a method for extracting human-interpretable rules from convolutional neural networks.
Scalable design patterns for multi-agent orchestration, reasoning workflows & secure enterprise integration.
Evaluating & deploying Azure OpenAI, Vertex AI / Gemini Enterprise & Databricks AgentBricks at scale.
M.S. in AI (Thesis Track), UNT. Published in MDPI on interpretable rules from CNNs.
Onsite engagements across 10+ countries spanning the US, Europe, the Middle East & Africa.
My current frontier at Tredence — agentic platforms, code intelligence, and AI across the whole development lifecycle.
A spec-driven meta-platform that spawns agents on demand — deliberately agnostic across cloud, framework, vector DB, runtime and LLM. You describe the agent in a spec; the platform assembles, wires and ships it. Swap any layer — model, runtime, store — without a rewrite, and no vendor lock-in.
Ingesting entire code estates into a knowledge graph over code — capturing inter-repo and system-wide dependencies to predict blast-radius and automate the SDLC.
Bringing AI across the full development lifecycle — accelerating greenfield builds and modernizing brownfield legacy systems with the same agentic tooling.
Production IDP combining Azure Content Understanding, Databricks AI Parse & AI Extract and fine-tuned SLMs for accurate, cost-efficient extraction at scale.
The right store for each job — vector and graph side by side: Azure Cosmos DB & Cosmos Gremlin, Postgres pgvector, and graphs on Postgres via Apache AGE.
Webinars, blogs & features on enterprise GenAI, Agentic AI and Intelligent Document Processing.
Watch the full webinar on building scalable, agentic Intelligent Document Processing (IDP) on Azure.
Watch on YouTube →Featured on the Microsoft Azure Cosmos DB engineering blog (March 2026).
Read on devblogs.microsoft.com → Tredence BlogA deep dive into building agentic IDP pipelines on Databricks AgentBricks.
Read on tredence.com → TM Forum · WebinarTM Forum webinar feature on scaling enterprise AI in telecom.
View on LinkedIn → Tredence · LinkedInTredence's feature on agentic AI work and the #BeyondPossible initiative.
View on LinkedIn →From the math of neural networks to enterprise-grade LLMOps — here's where I go deep.
Relative depth across the modern GenAI stack.
Where my day-to-day energy goes.
A sample of the GenAI / LLMOps stack I build with.
My Master's research on Explainable AI — extracting human-interpretable rules from CNNs.
MDPI — Advances in Explainable Artificial Intelligence, 2nd Edition
University of North Texas
University of North Texas, USA
Thesis on Explainable Deep Neural Networks (RICE). Built fully-connected nets for binary & multi-class text classification from scratch with custom penalties & loss functions. Reviewed 50+ top AI/ML papers.
Technical University, Rajasthan, India
Foundations in computer science & engineering.
Onsite engagements across the Americas, Europe, the Middle East & Africa.
Telecom data & revenue assurance → machine learning → enterprise GenAI architecture.
Defining enterprise North Star architectures for Generative & Agentic AI — scalable patterns for multi-agent orchestration, reasoning workflows, RAG knowledge systems & secure integration. Advising C-level stakeholders and driving large-scale deployment of Azure OpenAI, Vertex AI / Gemini Enterprise & Databricks AgentBricks. Currently building a cloud-, framework-, vector-DB- and LLM-agnostic meta-platform that spawns agents from specs, plus a code knowledge graph that automates the AI development lifecycle across greenfield & brownfield.
Led enterprise-wide unstructured-data handling for RAG: ingestion, chunking, enrichment, metadata extraction, topic modeling, embedding & indexing. R&D on multi-query retrieval (HyDE, intent extraction, hybrid keyword+semantic, re-ranking). Built the knowledge layer ingesting code repos to map system-wide impacts and automate the SDLC.
Moved an African telco from static-threshold to ML-based fraud detection with an open-source stack and an interactive GUI for real-time precision/recall trade-offs. Designed & deployed a North Star architecture that helped an AI startup scale 10× with reduced hallucinations. Sized GPUs and fine-tuned models from 130M to billions of params (VGG16, BERT-large, Nomic, Llama 3B).
Led engineers & consultants delivering real-time analytics on billions of daily records. Ensured 99.9% uptime, deployed ML models that averted $5M in fraud losses, and automated workflows suspending 200K+ fraudulent subscribers — AWS-native, Python & ML.
Led revenue & fraud management with real-time roaming fraud detection PoCs. Improved data integrity & governance and deployed SAS-based controls that prevented $400K in losses while hardening access to sensitive customer data. Python & Java; first ventures into AI.
Defined scope & strategy for Revenue & Business Assurance, built a master-data-management framework for subscriber data, delivered leakage KPIs, and reduced false positives for more accurate revenue reporting.
Deep data analysis to uncover new leakage & fraud scenarios, implemented system controls, expanded data coverage and stood up 24×7 monitoring dashboards enabling predictive maintenance.
Led global telecom analytics engagements across fraud detection, churn & network optimization. Designed metrics for high-revenue / high-risk zones with real-time sim-box controls and signaling-level auto-deactivations. Statistical auto-thresholding — my first exposure to learning algorithms.
Built production rules for the Mediation system processing billions of CDRs in near real-time with fail-safe delivery to multiple downstream systems.
Embedded & robotics project design and development for the Department of Science & Technology.
Led data migration for product upgrades, deployed & configured BSS stack (rating, charging, billing, mediation) and worked with smart memory databases (SMDB) for real-time rating & charging — state-of-the-art for its time.
Advising on enterprise GenAI strategy, Agentic AI architecture, RAG systems or LLMOps — or just want to talk Explainable AI? I'd love to hear from you.