Wind Pioneers
Local-first document intelligence for wind engineering
Backend & AI Infrastructure Engineer
I build reliable systems for AI and autonomous software.
Purdue CS senior with experience building production document pipelines, distributed ML infrastructure, backend platforms, developer SDKs, and verifiable agent systems.
Seeking 2027 new-grad opportunities in backend engineering, AI infrastructure, and platform engineering.
02Professional evidence
Solsten
98.2% extraction accuracy with controlled LLM fallback
Kong Labs
Distributed clinical-speech ML pipelines
BoilerBlockchain
Technical curriculum delivered to 100+ students
03Experience
Production and research engineering across document intelligence, distributed ML, backend platforms, and verifiable agent interactions.
Wind PioneersAI/ML Engineering Intern
Local-first document intelligence for confidential wind-engineering records.
- Built
- A hybrid PDF and table extraction pipeline using Docling, pdfplumber, local LLM inference, deterministic validation, confidence scoring, and human review.
- Hard problem
- Engineering values can span tables and multiple dependent documents, while confidential source material must remain local and auditable.
- Evidence
- Resolves cross-document dependencies and exports validated structured data into downstream engineering software formats without exposing private source content.
Python · Docling · pdfplumber · Local LLMs · Deterministic validation
SolstenSoftware Engineering Intern
Controlled LLM fallback for medical invoice extraction.
- Built
- A fast-path / slow-path pipeline that used deterministic parsing for high-confidence fields and Gemini fallback only for missing or ambiguous values.
- Hard problem
- Semi-structured invoices varied across layouts, while a pure-LLM approach was too expensive and a pure-regex approach was too brittle.
- Evidence
- Reached 98.2% field-level accuracy, reduced inference cost by approximately 65%, and reduced manual correction work by approximately 80%.
Python · pdfplumber · Vertex AI · Gemini 2.5 Flash · DSPy · pytest
Kong LabsUndergraduate Research Assistant
Distributed clinical-speech preprocessing and federated ML infrastructure.
- Built
- An end-to-end preprocessing and training workflow for clinical speech data, including containerized ingestion and low-latency feature extraction with Whisper and Librosa.
- Hard problem
- Distributed datasets required repeatable preprocessing and federated workflows that could operate across institutional boundaries.
- Evidence
- Processed 200+ clinical audio samples, increased preprocessing throughput by approximately 1.5×, reduced latency, and improved classification accuracy under distribution shift.
Python · PyTorch · Whisper · Librosa · Docker · Federated Learning
MOI LabsSoftware Engineer / Project Lead
Identity and traceability infrastructure for autonomous-agent interactions.
- Built
- TypeScript client and server SDK wrappers that attach identity and trace metadata to agent-to-agent requests and record verifiable request/response history.
- Hard problem
- Distributed agent interactions needed attribution, reliable fallbacks for unstable network state, and an end-to-end way to verify that recorded proofs matched execution.
- Evidence
- Led SDK development, designed ephemeral-state fallbacks, wrote end-to-end verification workflows, and co-authored the technical specification before exposing the MOI Chain implementation details.
TypeScript · Node.js · Google A2A Protocol · MOI Chain · Coco contracts
Leadership and ventures
Jan 2025 — presentBoilerBlockchain, PurdueHead of Education
Designed and taught a ten-week technical curriculum, improved completion rates, and led engineering education for more than 100 students.
2026 · OngoingManoCo-founder / Software Engineer
Co-founded and built product flows, backend services, and an AI workflow for coordinating commercial real-estate diligence.
May 2025 — PresentBoilerBite · Purdue SPARKCo-founder / Software Engineer
Replaced a Firebase prototype with a Node.js and PostgreSQL API using role-based access control, validated inputs, and a production Linux deployment.
04Selected systems
I work on three parts of reliable autonomous systems: dependable pipelines, controlled execution, and verifiable state.
Memora
Infrastructure for recording and verifying what AI agents did.
Memora records agent actions, authorization, inputs, outputs, identity, and execution history so teams can investigate failures and verify how a result was produced.
- Engineering decision
- Keep sensitive payloads encrypted off-chain while using signed execution receipts, ordered hashes, and explicit permission records as independently verifiable evidence.
- Evidence
- Published TypeScript and Node.js SDKs · 100+ contract tests · extensive Playwright end-to-end coverage
- Role
- Creator and lead engineer
TypeScript · Node.js · Hedera HCS · Hedera EVM · Solidity · ethers.js · IPFS · Supabase · AES-256-GCM · EIP-191 · pnpm workspaces
SmartInvoice
A hybrid extraction pipeline that uses deterministic parsing first and LLM fallback only when needed.
The system extracts semi-structured medical invoice fields, validates confidence and schema quality, and escalates only missing or ambiguous values to an LLM fallback.
- Engineering decision
- Treat the model as a controlled slow path behind deterministic extraction, validation, and evaluation rather than as the default parser.
- Evidence
- 98.2% field accuracy · 65% lower inference cost · 80% manual correction reduction
- Role
- Software Engineering Intern · Solsten
Python · pdfplumber · Gemini 2.5 Flash · Vertex AI · DSPy · pytest
Clinical Speech ML Pipeline
Clinical-speech preprocessing and federated training infrastructure for distributed datasets.
The pipeline containerizes ingestion and feature extraction, prepares clinical audio for repeatable training, and supports federated workflows across distributed data boundaries.
- Engineering decision
- Standardize preprocessing and packaging before federated training so model evaluation is not confounded by inconsistent data preparation.
- Evidence
- 1.5× preprocessing throughput · reduced latency · improved classification accuracy under distribution shift
- Role
- Undergraduate Research Assistant · Kong Labs
Python · PyTorch · Whisper · Librosa · Docker · Federated Learning
05Selected writing
Notes on evaluation, agent isolation, and the engineering boundaries that make autonomous systems easier to investigate.
01Retrieval Adaptation PipelineA 3,292-example procedural disruption benchmark from YouCook2 and WikiHow, and a config-driven evaluation pipeline covering retrieval augmentation, reranking, and VLM strategies — with persisted retrieval and generation traces for reproducible multi-method comparison.02AgentOSIsolated git worktrees, command allowlists, diff inspection, and a Rust-backed desktop runtime — bounded coding-agent execution with human review controls. A browser demo shows the UI and event flow; the desktop Tauri runtime handles real worktree operations.03TrustGraphFour-predicate deterministic scanning for trust-boundary vulnerabilities — executable Foundry PoC exploit tests, Markdown/JSON reports, and VS Code diagnostics. The LLM enriches explanations; it never creates or suppresses a finding.
Seeking 2027 new-grad opportunities in backend engineering, AI infrastructure, and platform engineering.
I’m especially interested in teams building backend platforms, production AI infrastructure, developer systems, distributed runtimes, and tools for operating autonomous software reliably.
Purdue CS senior · Graduating 2027