Skip to content
VDAI with VD

Lead AI Engineer at WeyBee Solutions

AI systems and reference architectures built for real constraints

I am Vishvdeep Dashadiya. I design, build and deploy agents, retrieval systems, fine-tuned models and the MLOps around them, and publish independent product research where the design itself is the evidence. Client delivery, self-built systems and target reference architectures are labelled separately. Evidence is documented and labelled as measured results, demonstrated capabilities, or targets. Agentic AI, real-time ML systems, and cloud-native infrastructure.

Book a callView case studies
7+ years in production AIAgents, RAG, fine-tuning, MLOpsNDA-friendly, documented evidence

Built with

  • Python
  • FastAPI
  • LangGraph
  • LangChain
  • Claude
  • Anthropic
  • OpenAI
  • Gemini
  • Mistral
  • Llama
  • Hugging Face
  • PyTorch
  • Ollama
  • AWS
  • Docker
  • PostgreSQL
  • Redis
  • Milvus
  • Qdrant
  • Neo4j
  • Celery
  • LiveKit
  • Next.js
  • React
  • TypeScript
  • Vercel

Selected work

Founders, self-built products and independent research

Client delivery, self-built projects and independent product R&D, with measured results, demonstrations and target designs clearly distinguished.

Self-Built Project

Measured

Svarupa: Verified Architecture Diagrams from Real Code

A Python CLI that renders architecture diagrams where every node and edge carries file:line evidence, plus a churn-free lockfile that puts architecture diffing into every PR.

Measured outcomes

Numbers are from the project's own test suite and validation runs on a 592-module production backend; no comparative benchmark against other tools is claimed.

856 tests

Gate every commit: unit, determinism, mutation harnesses

0 lines

Median architecture.lock churn per commit, measured on real history

Read case study

Independent Product R&D

TargetTarget Reference Architecture

Designing Secure Agent Transaction Infrastructure

A protocol-neutral reference design for giving AI agents bounded authority, exact approvals, isolated credentials, safe outcome handling and auditable transaction evidence.

Design outputs

The counts below describe elements in this target reference design and its public documentation; they are not observed production results or performance measurements.

7 controls

Risks, enforcement points and validation methods mapped

4 identities

Human, agent, workload and downstream credential kept distinct

Read case study

hypREspace

Measured

hypREspace: Cutting Token Costs 90% While Raising Tool-Calling Accuracy to 95%

Dynamic prompt injection, a gold evaluation set and Amazon Nova fine-tuning cut a live agent's token spend by 90% and lifted tool-calling accuracy from 75.8% to 95%.

Measured outcomes

Measured with before-and-after per-query token counts and tool-call accuracy scored against the gold evaluation set.

90%

Token cost reduction, measured per query

75.8% to 95%

Tool-calling accuracy on the gold eval set

Read case study
All case studies

What I do

Services built on shipped work

Eight services. Each links to the case study or project that proves it.

MLOps

Pipelines, monitoring, versioning and cloud deployment for models in the wild.

RAG Applications

Retrieval systems with hybrid search, reranking and traceable answers.

Voice AI

Real-time voice assistants with explicit local and online processing boundaries.

Process

How I work

Four steps, the same on every engagement, so you always know what happens next.

  1. 01

    Discover

    I interview the people who own the problem and look at real data before proposing anything.

  2. 02

    Specify

    A written specification and architecture, with the evaluation that defines "working", before code exists.

  3. 03

    Build with evaluation

    Small reviewed steps on a preview deployment. Every behaviour change is scored, not eyeballed.

  4. 04

    Ship and monitor

    CI, rollback and monitoring for cost, latency and quality, then a handover you can actually use.

Vishvdeep Dashadiya, Lead AI Engineer

Behind AI with VD

Vishvdeep Dashadiya, Lead AI Engineer

I design, build, and deploy intelligent products with rigorous engineering, clear metrics, and a calm delivery cadence. From agentic workflows to resilient MLOps, we move from prototype to scale without chaos.

7+ yrs

Product and platform delivery

GenAI

Agents, retrieval, copilots

MLOps

Monitoring, deployment, scale

Credential

Board of Studies, External Industry Expert

Atmiya University, since May 2024

Worked with

  • WeyBee Solutions
  • BankBenchers
  • hypREspace
  • 7Span
  • fxis.ai
  • Accenture
About me

Selected projects

Self-built systems and independent product R&D

Working software and target reference designs, with evidence and public-code or request-only source access clearly labelled.

Svarupa

Architecture diagrams that carry their own evidence

Self-Built ProjectMeasured

A Python CLI that reads a codebase with tree-sitter (pinned Python and TypeScript/JavaScript grammars) and produces verified architecture diagrams plus a queryable knowledge graph. Every node and edge carries file:line evidence or it does not render; omissions are stated as diagnostics, never fabricated. Published on PyPI under the MIT license.

Pythontree-sitterNetworkX

Secure Agent Transaction Infrastructure

Bounded authority for agents that can transact

Independent Product R&DTargetTarget Reference Architecture

Tool access does not by itself establish transaction authority. This protocol-neutral target reference architecture turns agent proposals into bounded, policy-checked, transaction-bound, credential-isolated and auditable external actions.

Deterministic PolicyTransaction-Bound ApprovalScoped Grants

Adaptive RAG

Toggle-Driven Hybrid Retrieval With Skills

Self-Built ProjectDemonstrated

A self-hosted LangGraph and FastAPI RAG service for a regulated finance and legal practice. Vector search always runs; allowlisted web search is added only when the request toggles it on, never by an LLM router. Four self-reflection gates with hard-capped retry loops decide whether an answer ships. Slash-invocable skills run as isolated sub-agents over the same graph, and opt-in Langfuse tracing plus async DeepEval scoring observe every run without adding latency.

LangGraphFastAPIMilvus
All projects

Writing

Latest writing

Production lessons and independent reference-design research across agents, retrieval, fine-tuning and MLOps.

September 25, 2026 · 7 min read

Why I Built Svarupa: Architecture Diffing in Every PR

A clean-looking PR rerouted around a load-bearing layer and the diff didn't show it. The lockfile design, --diff, --drift-base, and CI setup that came out of that review.

ArchitectureCI/CD

September 25, 2026 · 6 min read

Using Svarupa: From Clone to Verified Map in Minutes

Install svarupa, scan a repo, and read the evidence-backed report: artifact contents, the viewer, the query CLI, and the MCP server — anchored on two live demo reports.

Developer ToolsTutorial

September 25, 2026 · 6 min read

Svarupa for Agents: Give Your Coding Agent a Verified Map

Your agent re-derives the architecture from grep on every task, and guesses confidently when the fragments lie. Hand it the verified graph instead — over MCP, with file:line citations.

MCPAI Agents
All posts

Questions people ask before we start

Are you a company or a person?

A person. I work as Lead AI Engineer at WeyBee Solutions, and AI with VD is my independent practice. You work with me directly, and I bring in specialists only when a project needs them and you agree.

What kind of clients do you work with?

Founders and product teams at startups and small enterprises who need production AI, from a first agent to a fine-tuned model in a live product.

How do engagements start?

With a call, then a short written proposal: problem, approach, deliverables, timeline. Nothing starts until we both agree on what "done" means.

Do you sign NDAs?

Yes. Several case studies on this site name the client only as an anonymous label because of an NDA.

Which stack do you use?

Python, FastAPI and LangGraph on the backend, TypeScript and Next.js on the frontend, PostgreSQL, Redis, Docker, and AWS. Open models where they fit, hosted models where they win.

Can you work with my existing team?

Yes. Most engagements plug into a team as the senior engineer for the AI parts, with code review and documentation as part of the job.

Ready to build something that holds up in production?

Tell me about it. I reply within one working day with a first take and no sales pitch.

Book a callSend a message