Work - Priyatham Chimmani

W-09 · Independent build · Public data

17units in a synthetic 3D plant

Plant Blueprint, a Synthetic Digital Twin

Problem: Public chemical-plant statistics are tables; it is hard to see which figure can honestly describe which kind of unit.
What I did: Built an interactive 3D blueprint of a generic plant with a seeded simulation whose incident rates come from committed EPA RMP and OSHA outputs, plus Foundry-style alerts, actions and an audit log.
Status: Live. A synthetic site, not a real facility; no live telemetry, behaviour is a model and every statistic cites its source file.
Stack: three.js, JavaScript, Python

W-08 · Independent build · Public data

7ontology object types

Problem: A pipeline shows that a few implausible OSHA filings can move the national injury rate; an EHS team needs to act on that per site and per filing.
What I did: Designed a Foundry-style ontology: linked object types, governed analyst actions with an audit log, and functions that reuse the tested pipeline.
Status: Merged and tested on a synthetic fixture; the real-data path has not been run. Not affiliated with Palantir; no employer data.
Stack: Python, pandas, JSON spec, SQLite

W-07 · Palantir Foundry · In review

12ontology object types

Problem: Cases, events, concerns and insights sat in separate lists, and each report could count them differently.
What I did: Modelled them as 12 linked Foundry object types, put every KPI and status rule in one logic module, and built a Gensuite-style widget on top.
Status: Verified on a Foundry branch over 190 cases, 1,221 events, 11,546 concerns and 82 insights; merge proposal in review.
Stack: Palantir Foundry Ontology, TypeScript, React

W-03 · Platform

80+enhancements shipped

Problem: Every global site captured safety data through an aging interface, each with its own workaround.
What I did: Led a two-phase refresh of Hexion's enterprise EHS application, from requirements through global rollout.
Outcome: 80+ feature enhancements; every global site now captures safety data through the refreshed version.
Stack: Gensuite (Benchmark ESG)

W-02 · Award winner

Transformation Awardadopted company-wide

Problem: No shared model for comparing competitive opportunity across dozens of micro markets.
What I did: Built a Python linear-programming engine that scores and ranks those markets from an explicit objective and constraints.
Outcome: Adopted company-wide and awarded the Transformation Award.
Stack: Python, linear programming

W-04 · AI

3agent platforms, one grounding layer

Problem: An agent answering freely from live systems can produce a fluent, wrong safety number.
What I did: Built the MCP architecture that grounds NLQ and summarization agents in live systems, with deterministic code for exact calculations.
Outcome: Agents in production on Copilot Studio, Claude Code and Codex, with an audit trail behind every answer.
Stack: MCP, OpenAI API, Copilot Studio, Claude Code, Codex

Go deeper