Pranava Jayanthi Krishna

Senior Software Engineer. Dallas, TX.
I build
agents that plan queries instead of guessing them

Most of what makes an agent trustworthy is not the model. It is the query federation, the isolation, and the orchestration underneath. I build that layer, and the product on top of it.

Patent

Chat-interface for natural language driven data management

A system for connecting to data sources, generating ranked execution plans for natural language questions, and presenting those plans to users for modification.

In plain terms: ask a question in English, see how the system intends to answer it, and change the plan before it runs.

Stage 01 of 03Stage boundary

Plume Design, Inc.

Senior Software Engineer
April 2026 to presentDallas, TX
04

Multi-tenant SQL over the data lake

node 04 · stage 01, Plume Design

The query execution layer inside Prism: an ISP partner runs SQL against a shared analytical store and only ever sees its own data. The service is a broker, not a query rewriter: identity comes from the authenticated session, and every query runs under a short-lived authority derived from it.

constraint
A partner may only reach its own databases and tables, never another's. Enforced by Lake Formation, not by the API adding a filter to each query, so forgetting a step returns nothing, not someone else's rows.
TypeScriptAthenaLake FormationS3
Stage 02 of 03Stage boundary

Promethium

Senior Software Engineer
April 2023 to March 2026Dallas, TX
11

Agentic system over federated queries

node 11 · stage 02, Promethium

Automode turned a plain-English business question into federated SQL across a customer's sources. I worked on the pivot that put generative models behind its existing NLP, reasoning and SQL-building stages, one stage at a time, with the old path standing beside the new until it earned its place.

answers accepted by the askerunder 25% → 70%
usage over the same windowroughly 9×
PythonLangChainpgvectorTrinoJava
Stage 03 of 03Stage boundary

Accenture

Software Engineer
September 2020 to August 2022Bengaluru, Karnataka
17

Client-facing SaaS for workflow assessments

node 17 · stage 03, Accenture

Owned the product end to end: the ASP.NET Core REST API in C#, the PostgreSQL data layer, the Angular front end. A consultant worked through a client's operations question by question; the system ranked which processes were worth automating, with every recommendation traceable to the answers that produced it.

constraint
Customer requirements varied per deployment, so the frontend had to be assembled, not rewritten. Built as a library of dynamic, reusable Angular components.
C#.NET CoreTypeScriptAngularPostgreSQL
Education
University at Buffalo
M.S. Computer Science and EngineeringUniversity at Buffalo (SUNY)
October 2022 to January 2024Buffalo, NY
PES University
B.Tech Electrical and ElectronicsPES University
August 2016 to August 2020Bengaluru, Karnataka
Contact

The plan above is current, not finished. If you are building the stage that comes next, I would like to hear about it.

Dallas, TX