Four builds. The first two are platforms delivered inside enterprise constraints — described by architecture and outcome, with client detail withheld. The last two are mine end to end, and you can run them.
01Valtech · 2025 · Architect
CodeIQ
Valtech · 2025 · Architect
CodeIQ
Agentic .NET Legacy Modernization Platform
A hybrid agent stack that reads a legacy .NET estate, plans its migration, executes it, and proves the result still behaves.
Legacy modernization fails for a boring reason: nobody can hold a 200K-line codebase in their head, and neither can a language model. CodeIQ solves the context problem before it tries to solve the migration problem.
Six specialised agents that validate a lakehouse migration table by table, turning a five-day manual review into one day.
Migrating a data platform is the easy half. Proving the migrated data is correct — row counts, type fidelity, transformation logic, referential integrity — is the half that consumes person-years.
Natural-language product discovery that searches across multiple e-commerce sources and returns one merged result set.
Ask for "a black midi dress under ₹3000 that isn't polyester" and you get results, not a filter grid. An agentic workflow translates that intent into structured database and search queries, fans them out across sources, and reconciles the responses into a single ranked interface.
A query router that classifies intent and dispatches to either a fully local RAG pipeline or live web search.
One router, two very different retrieval strategies. The classifier decides whether a question is answerable from private documents or needs the live web, then dispatches to a local RAG pipeline or the Tavily API accordingly.
Five things, each with the receipt attached. If a line below can't be backed by something I shipped, it isn't on the list.
Designing agent systems as organisations rather than prompt chains: explicit role decomposition, handoff contracts between roles, and a shared context surface each agent reads from instead of re-deriving. The design goal is attributable failure — when a run goes wrong, you can name which role got it wrong.
Evidence
Architected a four-role subagent layer (Architect, Developer, Lead, Testing) driving end-to-end modernization of 160K–200K LOC .NET Framework repositories.
Claude Code Subagents
Google ADK
LangGraph
LiteLLM
EXPERIENCE
Timeline
How I got here.
One employer so far, and a fast eighteen months inside it.
Apr 2025 — Jul 2025
Apr 2025 — Jul 2025
AI Engineer Intern
Valtech · Bengaluru
Internship at Valtech.
Sep 2025 — Present
Sep 2025 — Present
AI Engineer
Valtech · Bengaluru
Architecting agentic platforms for legacy modernization, documentation and data-migration validation.
CONNECT
Contact & reach out
Get in Touch
Hiring for an agentic-AI or LLM-systems role, or want a second opinion on an architecture before you commit to it? Reach out directly — email gets the fastest reply.