AI & Data Solutions Architect
Agentic AI that passes the audit.
I’m Yogendra Bhardwaj. I architect production generative AI and multi-agent systems for regulated life-sciences and financial-services enterprises — grounded, evaluated, and audit-ready from day one, built on 18 years of data engineering at Accenture.
- GxP
- HIPAA
- FDA 21 CFR Part 11
- PCI
- agent.planAGENT
- tool.graphrag.searchTOOL
- tool.assay.queryTOOL
- guardrail.pii_maskGUARD
- llm.synthesizeLLM
- eval.groundednessEVAL
- human.approveHITL
- audit.logAUDIT
Illustrative trace — the kind of run I design for.
- Enterprise delivery at Accenture
- 18yrs
- Service reps on a grounded RAG assistant
- 200+
- Sprint planning with an MCP-driven agent
- 15→5days
- Faster build cycles via AI forward engineering
- 50%
- Users on a core banking platform I led
- 5,000+
- Microsoft Azure certified
- 4×
- Azure AI Foundry
- Anthropic Claude
- Microsoft Discovery Studio
- Azure OpenAI
- Databricks
- Model Context Protocol
- GitHub Copilot
- Copilot Studio
In plain words.
I help large, regulated companies — drug makers, banks and credit unions — use AI safely. I design the systems where AI assistants and agents find the right information, take actions with software tools, and stay inside the rules, with people approving the decisions that matter.
Before AI, I spent 15 years at Accenture building data platforms, cloud systems and large migrations, so I know the plumbing that AI depends on — and how to get it through an audit.
Outside client work, I teach: guest lectures and hands-on sessions for college students on AI, data and careers in technology.
- Role
- AI & Data Solutions Architect, Accenture
- Experience
- 18 years (2008 – present)
- Industries
- Pharma & life sciences · Financial services · Public transport
- Focus
- AI agents · Grounded AI assistants (RAG) · Data engineering · AI governance
- Certified
- Azure Solutions Architect Expert + 3 Azure certifications
- Based in
- Wilmington, Delaware, USA
- Open to
- AI architecture leadership roles, advisory work and speaking
Three layers of AI that holds up in production.
Multi-agent systems with humans in the loop
Specialized agents that plan, call tools and hand off to one another — with a person approving every decision that matters, and every tool call written to an immutable audit log.
- Azure AI Foundry Hosted Agents and Claude via the Anthropic API
- Action-based tool calling and custom tool-integration frameworks
- Model Context Protocol (MCP) servers for enterprise systems
- Execution guardrails, strict RBAC, and approval gates
- Planner agent
- Tools · MCP
- Guardrails
- Human approval
- Audit log
# agent policy — illustrative
agent: target-validation
model: claude # via Azure AI Foundry
tools:
- graphrag.search # literature + pathways
- assay.query # read-only, RBAC-scoped
guardrails:
prompt_filter: on
tool_allowlist: strict
human_in_the_loop:
required_for: [hypothesis.promote, dataset.write]
audit:
every_tool_call: immutableAnswers that cite their source — or decline
Hybrid retrieval over documents and structured data, evaluated against a golden test set before every release. When retrieval is weak, the assistant says so and routes to a person instead of guessing.
- Hybrid vector + structured retrieval; GraphRAG knowledge stores
- Groundedness, relevance and coherence evals as a release gate
- Citations on every answer; refusal routing on low confidence
- PII masking and content-safety filters before the model sees data
- Query
- PII mask
- Hybrid retrieve
- Grounded answer
- Eval gate
# release gate — illustrative
scores = evaluate(golden_set, metrics=[
"groundedness", "relevance", "coherence",
])
if scores.groundedness < THRESHOLD:
block_release("groundedness regression")
# at runtime: decline, don't guess
if retrieval.confidence < MIN_CONFIDENCE:
return route_to_senior_staff(query)
return answer_with_citations(query, docs)The data foundations AI actually runs on
Fifteen years of pipelines, platforms and migrations before GenAI — and now GenAI pointed back at legacy estates to reverse-engineer them into specifications.
- Azure Data Factory, Databricks, Synapse, Data Lake, Snowflake
- IoT ingestion and predictive-maintenance analytics
- Legacy-to-cloud migration under 21 CFR Part 11
- AI-assisted reverse engineering of legacy codebases
- Legacy / sensors
- Ingest · ADF
- Databricks
- Data Lake
- BI · RAG
# AI reverse engineering — pipeline sketch
legacy repo
→ chunk by module / call graph
→ Claude + Copilot: explain, trace, summarize
→ functional spec + technical spec
→ analyst review → versioned in source control
result: 5+ legacy apps documented, −30% analyst effortCase studies from regulated delivery.
Client names withheld. Architectures described generically.
Governance is a design input, not a review step.
- 01
Guardrails first
Safety filters, tool allowlists and PII masking are designed in on day one — not bolted on before the audit.
- 02
Eval before every release
A curated golden set and automated groundedness checks gate each deploy, so quality regressions never reach users.
- 03
Audit trail by default
Every agent tool call is logged immutably. If a regulator asks “why did it do that?”, there is an answer.
- 04
Humans stay in the loop
Agents propose; people approve. Low-confidence answers route to an expert instead of guessing.
Bringing industry AI into the classroom.
Guest lectures and hands-on sessions for college students on generative AI, agents, data and cloud careers.
- 01Generative AI in the enterprise: from prompt to production
- 02Building AI agents with tools and the Model Context Protocol
- 03RAG that doesn’t hallucinate: evaluation and guardrails
- 04Responsible AI in regulated industries (GxP, HIPAA, PCI)
- 05Careers in data engineering and cloud architecture
- 06From legacy to cloud: modernization with AI assistance
From .NET to data to agents.
Eighteen years at Accenture — every layer of the stack that today’s AI depends on.
Full résumé →- Feb 2026 – Present
Agentic AI Technical Architect
Life Sciences · Drug Discovery
- Oct 2023 – Jan 2026
Solution Architect & AI Delivery Lead
Financial Services · Credit Unions & Banking
- Apr 2019 – Sep 2023
Technical Architect, Clinical Trial & eDocs
Life Sciences · Clinical Operations
- Aug 2015 – Mar 2019
Data Engineer & Application Development Lead
Public Transportation · Asset Management
- Jan 2012 – Jul 2015
Clinical Application Migration Lead & SharePoint Architect
Clinical & Enterprise Platforms
- Apr 2008 – Dec 2011
Sr. Software Engineer & Team Lead
Enterprise Application Development
Building agents for a regulated business?
I’m happy to talk architecture, delivery, speaking invitations, or roles where governed AI matters.
