Manish Kumar
← All work

agentic ai

Agentic AI Platform

Internal production platform · Tru

Agent workflows at Tru: retrieval, multi-step tasks, and the reliability work.

Software Engineer·2024 – Present·Mohali

Enterprise · NDA

Context

At Tru, I work on an internal agent platform used for multi-step workflows across existing systems.

Problem

Teams needed agents that could retrieve context, run tasks in sequence, and talk to the APIs they already had.

Constraints

  • Production stability and observability requirements
  • Vector retrieval latency and accuracy trade-offs
  • Multi-step agent state management across failures

My contribution

  • Implemented FastAPI services supporting AI-driven agent workflows
  • Integrated Weaviate for vector retrieval and PostgreSQL for persistent agent state
  • Built agent execution logic with LangGraph and LangChain for task orchestration
  • Shipped frontend surfaces in React/Next.js for monitoring and interaction

Solution

A layered architecture: FastAPI API layer → LangGraph agent graphs → Weaviate retrieval + PostgreSQL persistence → React dashboards for operators.

Results

  • Enabled production agent workflows with retrieval-augmented context
  • Reduced manual task handling through orchestrated multi-step agents
  • Improved system stability through production debugging and code reviews

Stack

FastAPIPythonLangGraphLangChainWeaviatePostgreSQLVector DBReact

Working on something like this?

Write to me if the work looks close.