Dishant Shah
Newark, CA | dishants560@gmail.com
LinkedIn: linkedin.com/in/dishants560 | GitHub: github.com/Dishah3241
Software Engineer — Agentic Evaluation & AI Developer Tooling
Software engineer with 5+ years at Intapp building backend services, data pipelines, and cloud infrastructure, now focused on agentic evaluation tooling — built an MCP server and connectors for a central eval platform and drove its adoption across at least 4 teams. Previously led an AWS-to-Azure consolidation of 30 microservices into 3 AKS mini-monoliths. Independently, directs AI-agent formalization projects in Lean 4; two are linked from Google DeepMind's formal-conjectures repository.
Experience
Software Engineer, Intapp — Palo Alto, CA
August 2021 - Present | Cloud-based SaaS platform for professional services firms (legal, accounting, finance)
- Build agentic evaluation pipelines; the team's central eval tooling was adopted by other teams that had previously run disjointed, uncoordinated eval work, with onboarding support through discovery sessions, platform demos, and connector walkthroughs.
- Built the platform's MCP server and connectors, originally for the core team's use; the MCP became the onboarding lever that made cross-team adoption practical.
- Helped implement standardized pathways for model updates and new-model evaluations, so teams can adopt the newest models through a consistent eval process.
- Led migration from AWS microservices to Azure AKS, consolidating 30 microservices into 3 mini-monolith Docker images, reducing operational overhead and network latency across service boundaries.
- Implemented a caching layer for mostly-static data, reducing API response times from ~2 minutes to ~3 seconds (~97%) by eliminating redundant database round trips.
- Proposed and executed architecture refinements to eliminate unnecessary network hops, reducing end-to-end update times from 30 seconds–several minutes to under 2 seconds.
- Also: refactored database schemas and simplified complex queries; built an SMTP email delivery server from scratch in Node.js; migrated data delivery pipelines from real-time Kafka (AWS) to batch polling (Azure), with Grafana/Prometheus dashboards and automated test suites through the migration.
- Manage cloud infrastructure across AWS (Lambda, CloudFormation, Terraform) and Azure (AKS, Argo CD, Helm) via Jenkins CI/CD; built CI/CD pipelines for PySpark data applications migrated from Databricks to AKS.
Open Source / Projects
Formalization in Lean 4 — graph dimension & Erdős problem 1007
Sep–Oct 2026 | Lean 4, Mathlib | github.com/Dishah3241
- Directed AI coding agents (Claude, Grok, GPT via Codex, GLM) to formalize published results on the Erdős–Harary–Tutte dimension of graphs, building a reusable Lean 4 library (GraphDimension) and six proof repositories with no
sorry, kernel- and NanoDa-comparator-checked proofs, axiom-audit CI, and leanblueprint documentation.
- Formalized a known result — Chaffee & Noble's dimension-five bound for Erdős problem 1007 — and the dimension-four extremal case; both formalizations are linked as
formal_proof references from Google DeepMind's formal-conjectures repository (PR #6511, merged Sep 23, 2026; PR #6623, merged Sep 27, 2026). The formalization is of a published proof, not a new mathematical result, and was written by AI agents under my direction, as disclosed in both PRs.
agent-stream — observability for headless coding agents
Oct 2026 | Bash, jq, Agent Client Protocol | github.com/Dishah3241/agent-stream
- Tooling that renders and captures event streams from headless coding agents (Claude Code, Pi, Cursor, Grok, Codex, ACP agents) as narrated terminal output; every run leaves a complete on-disk record — raw events, stderr, rendered display, and a structured
state.json other programs can read mid-run — plus a fleet watcher and a multi-machine board over SSH.
pve-watch
Oct 2026 | Go, Bubble Tea | github.com/Dishah3241/pve-watch
- Read-only terminal UI for a Proxmox cluster (nodes, VMs/LXCs, storage, and a problems lens); designed to never send a write to the Proxmox API.
Skills
Languages: Python, TypeScript/JavaScript, Go, Lean 4
Backend & Data: Node.js, FastAPI, REST APIs, PySpark, PostgreSQL
Cloud & Infrastructure: Azure AKS, AWS, Kubernetes, Docker, Kafka
DevOps & Observability: Jenkins, Argo CD, Helm, Terraform, CloudFormation, Grafana, Prometheus
AI Tooling: MCP, Claude Code, agent evals, multi-harness agent observability
Education
San Jose State University, San Jose, CA
Master of Science in Software Engineering — May 2022
University of Alabama at Birmingham, Birmingham, AL
Bachelor of Sciences in Biomedical Engineering, Minor: Computer Science, Chemistry — August 2018
First author, J. Clin. Med. (2020); 3 peer-reviewed papers (ACS Biomaterials Sci. & Eng. 2017; Nano Convergence 2016).