Wenhao HeJacky
Software Engineer · New York

Wenhao He

I build AWS-backed AI products that hold up in production, and I measure whether they do.

Engineer and founder at ResumeMatch, a live AWS product I designed, shipped, and operate solo, where a multi-pass Bedrock pipeline scores and rewrites résumés and a content-hashed cache answers repeats in 20ms. Before that, two years shipping full-stack web and mobile features at NYC startups.

Selected workRésuméGet in touch
Wenhao He
Role
SWE & Founder, ResumeMatch
Certified
AWS Developer, Assoc.
Education
M.S. AI · UB (SUNY)
Based in
New York, NY
20ms
Cached analysis, down from a 34s median across 140 requests
ResumeMatch
71%
Of AI rewrite edits proposed keywords the résumé never supported — caught offline, guarded at generation
ResumeMatch
2.9%
Unsupported claims, down from 6.4% on a frozen 32-case baseline
Financial Document Agent
44
Contrast failures a WCAG re-run caught, traced to five design tokens
ResumeMatch · 508 / 2.1 AA
01

About

Get to know me
about / wenhao he
Focus
AWS-backed AI products, LLM validation, and full-stack systems.
Currently
Software Engineer & Founder at ResumeMatch, New York.
Education
M.S. Artificial Intelligence and B.S. Computer Science, University at Buffalo (SUNY).
Interests
Production AI systems, LLM evaluation, RAG agents, event-driven backends, cloud cost optimization, and developer-quality tooling.
software_engineer.py
class SoftwareEngineer:

    def __init__(self):
        self.name = "Wenhao He"
        self.role = "Founder & Software Engineer"
        self._email = "wenhaohe8@gmail.com"

        self.education = [
            "MS in Artificial Intelligence, UB (SUNY)",
            "BS in Computer Science, UB (SUNY)",
        ]

        self.certifications = [
            "AWS Certified Developer, Associate (Feb 2026)",
        ]

        self.stack = {
            "languages": ["Python", "TypeScript", "Java", "Bash"],
            "frontend":  ["React", "Next.js", "React Native"],
            "backend":   ["Node", "GraphQL", "Kafka", "Spring"],
            "ai_ml":     ["Bedrock", "LangChain", "RAG"],
            "cloud":     ["Lambda", "DynamoDB", "S3", "Docker"],
        }

    @property
    def current_focus(self):
        return "AWS-backed AI products and full-stack systems"

    def __repr__(self):
        return f"{self.name} | {self.role}"
02

Selected work

One product, four builds
ResumeMatch
Live product

ResumeMatch

A fully serverless AWS product I designed, shipped, and operate solo. A six-pass Bedrock pipeline routes schema extraction, keyword extraction, scoring, and gap analysis to Claude Haiku and missing-keyword ranking and rewriting to Claude Sonnet, behind a content-hashed DynamoDB cache that cut repeat analyses from a 34s median to 20ms. An offline harness caught the rewrite proposing keywords the source résumé never supported in 71% of edits, so a deterministic evidence guard now drops them at generation time. Around it sit a voice-first mock interview on Deepgram Nova-3 with per-session keyterm prompting and Bedrock STAR feedback, and a Kanban outreach tracker that scores whether an application is worth chasing.

PythonReactAWS BedrockLambdaDynamoDBStripe
Visit ResumeMatch
Financial Document Intelligence Agent
AI · governed RAG agent

Financial Document Intelligence Agent

A governance-first ReAct agent answering questions over SEC filings and compliance policies on Claude Haiku via Bedrock, across a 10-handler document-format registry, with context-policy admission dropping chunks before the model sees them and a per-answer JSONL audit trail — 689 logged runs, 27% multi-hop. Then I caught my own answer-trust layer flagging correct answers: it scored claims against 700-char-capped audit excerpts, so figures deeper in a chunk read as unsupported. Fixing that moved a frozen 32-case baseline from 6.4% to 2.9% unsupported claims and 75% to 84% grounded.

PythonAWS BedrockLangChainChromaFastAPIReact
View on GitHub
Order Processing System
Backend · event-driven

Order Processing System

A Spring Boot event-driven order processing system with a transactional outbox, making order-event publication from PostgreSQL to Kafka reliable across service outages. Consumers are idempotent through a processed_events ledger with uniqueness constraints, so retries and redeliveries never execute an event twice.

JavaSpring BootKafkaPostgreSQLDockerKubernetes
View on GitHub
Audio Cloning on Historical Figures
Research paper

Audio Cloning on Historical Figures

A deep learning project cloning a judge's voice to narrate the decision from Brown v. Board of Education, using speech synthesis techniques for realistic voice generation. Written up with Dr. David Doermann at University at Buffalo.

TensorFlowPythonWaveNetSV2TTS
Read the paper
Monocular Depth Estimation
Research paper

Monocular Depth Estimation

An encoder-decoder CNN that estimates 3D distance from a single 2D image, with applications in robotics, autonomous vehicles, and augmented reality.

PyTorchComputer VisionCNNOpenCV
Read the paper
03

Experience & education

Nov 2025 - Present

ResumeMatch

Software Engineer & Founder · New York

Built and operate resumematchapp.com solo on AWS serverless — Lambda, API Gateway, DynamoDB, S3, and CloudFront — with Cognito auth and per-plan usage limits enforced in the Lambdas rather than the client.

Audited the frontend against Section 508 / WCAG 2.1 AA and published a conformance report stating its own limits. A re-run caught 44 untested contrast failures in the light theme, traced them to five design-token values, and cleared all five audited routes in both themes.

Cut repeat analyses from a 34s median to 20ms with a DynamoDB cache (SHA-256 content hashing, gzip), measured across 140 production requests, and zeroed model cost on hits.

Caught the AI rewrite proposing keywords the source resume did not support in 22 of 31 edits — 71% — using an offline harness I built (10 resumes, 3 configs, 60 runs). Shipped a deterministic evidence guard that drops those at generation time; 26 rejections in production.

Ship through GitHub Actions, which gates every push on lint, tests, and build before the deploy job runs, then syncs to S3 and invalidates CloudFront over GitHub OIDC with no stored AWS keys.

Shipped a Mock AI Interview feature on Deepgram Nova-3 speech-to-text with per-session keyterm prompting derived from the resume and job description, plus Bedrock per-turn STAR feedback and an end-of-session assessment.

Issued Feb 2026

AWS Certified Developer, Associate

Amazon Web Services
Sep 2024 - Jun 2025

Clipp Inc.

Full-Stack Software Engineer · New York

Eliminated N+1 queries and redundant frontend round-trips across ordering workflows by redesigning GraphQL schema and resolvers to batch nested order data.

Built a modular Jest test framework spanning Payments, Purchase Orders, and Returns, using strict TypeScript contract validation to catch cross-service data mismatches before release.

Fixed an invoice refund flow in the Next.js web app using shallow routing to keep URL query parameters in sync without remounting the page, inside a Yarn workspaces monorepo.

Apr 2024 - Sep 2024

CAN International Corp.

Full-Stack Software Engineer · New York

Removed manual data entry from the document intake pipeline by integrating Azure Form Recognizer, converting unstructured PDF job descriptions into validated typed JSON records.

Built and launched a React Native mobile app on Firebase and Azure supporting rewards, payments, and job board workflows.

Aug 2022 - Jun 2024

M.S. Engineering Science (Artificial Intelligence)

University at Buffalo, SUNY
Aug 2018 - Jun 2022

B.S. Computer Science

University at Buffalo, SUNY
04

Stack

What I work with
Languages
Python · TypeScript · JavaScript · Java · Bash
Frontend
React · React Native · Next.js · Tailwind · HTML/CSS
Backend
Node.js · Express · FastAPI · REST APIs · GraphQL · Microservices · Spring Boot · Kafka
Cloud & Data
AWS Lambda · API Gateway · DynamoDB · S3 · Cognito · CloudFront · CloudWatch · Bedrock · Azure · PostgreSQL · MySQL · MongoDB · Firebase
AI & ML
LLMs · RAG · AI agents · Embeddings · Vector databases · Model evaluation · Prompt engineering · LangChain · Chroma
DevOps & Testing
Docker · Kubernetes · Git · GitHub Actions · Agile · Jest · Playwright · pytest · Claude Code · MCP · Copilot
05

Writing

blog.wenhaohe.com
Latest post7/7/2026
When Real Technology Meets Bad Investment

Deep dives into machine learning fundamentals and system design patterns, from neural networks to transformers, and from video streaming to ticketing platforms.

Machine learning
System design
Engineering solutions
Read the blog
06 · Get in touch

Let’s build
something solid.

Open to software engineering roles in New York and remote, and always happy to talk about production AI, LLM evaluation, or event-driven backends.

LinkedInDownload résumé