Engineering systems that hold up under real-world constraints.

1. 01

   Adam

   GitHub ↗

   An agentic outreach pipeline that turns a fresh job posting into a personalized, verified, ready-to-send email — where LLM agents do the research and writing, and deterministic code owns every consequence.

   • 8-stage pipeline orchestrating 2 tool-calling Claude agents (contact research, drafting) — agents propose, typed & audited code decides, and each agent sees only the context it needs.
   • Human approval enforced by architecture, not policy: no code path can transmit a message — drafts land in Gmail over IMAP, backed by per-company rate limits, address verification and prior-contact checks against real sent mail.
   • LLM-as-judge qualification over 300+ live postings, with frozen anchor postings in every batch as a drift alarm — measured evals retired a legacy heuristic scorer the day it underperformed.

   • Python
   • Claude Agents
   • LLM-as-judge
   • SQLite
   • Hunter API
   • IMAP
2. 02

   Lunar Landing RL Hackathon 🏆 1st place

   GitHub ↗

   A reinforcement-learning agent that took first place at an RL hackathon by learning to land the LunarLander.

   • Deep Q-Network in PyTorch reaching a 285.3 reward score.
   • Finished 1st out of 50+ competing teams.

   • Python
   • PyTorch
   • Deep Q-Networks

client deployments owned end to end

15+

financial records processed daily

5M+

processing time cut

87%

headroom under current infrastructure

30×

1. Jun 2025 — now

   Software Engineer

   Finaptive

   The forward-deployed job, in practice: I run the discovery call, own the integration, and answer for it in production.

   • Own the full lifecycle of 15+ client deployments across SAP, Costpoint, Blackbaud, PointClickCare, Xero & SFTP — from discovery calls with CFOs through go-live and ongoing optimization.
   • Designed a production classification engine on the Claude API that maps unstructured client data to standardized financial schemas — the LLM proposes, deterministic validation decides what reaches reconciliation.
   • Led the PointClickCare marketplace integration end to end: technical validation calls, API safeguards, and a webhook-driven incident-alert system serving 10 healthcare facilities.
   • Architected a Celery/Redis dispatcher–executor pipeline — isolated queues, distributed locking for idempotency, rate-limit retries — cutting processing time 87%; handles 5M+ records/day on KEDA autoscaling with 30× headroom.
   • Built a connector framework (OAuth2, REST, ODBC, SFTP) that turned one-off integration work into zero-code client onboarding.

   • Python
   • Claude API
   • Celery
   • Redis
   • PostgreSQL
   • KEDA
   • Azure
2. Feb — May 2025

   Software Engineer Intern

   Finaptive

   Hired full-time four months in.

   • Built the initial ETL pipeline — transformation workflows, validation checkpoints, and Celery RedBeat scheduling for automated ingestion across client ERP environments.
   • Added data-preview endpoints so clients inspect upstream samples before full runs — catching schema issues early and cutting bad-transfer support tickets.
3. May 2023 — Apr 2024

   Design Automation Engineer Intern

   AMD

   • Built Python / C-Shell automation that eliminated 20+ hours of weekly validation work across hardware design workflows.
   • Led code reviews and static analysis across 10+ hardware design projects, surfacing 100+ critical defects.
   • Developed a real-time PHP/JS dashboard for test-result observability and system-risk triage.

Full résumé (PDF) ↗