AI & SOFTWARE DEVELOPMENT

Big ideas.
Intelligent
execution.

We turn ambitious ideas into practical AI solutions and thoughtfully built software. A little intelligence. A lot of possibility.

Good technology starts with a good conversation.

SCROLL TO DISCOVER

01 / WHAT WE DO

Built around your next move.

Complex challenges.
Clear solutions.

Technology should move you forward.
We help you find the right way to build it.

01

Practical AI.

Put intelligence where it makes a difference. Connect your knowledge, simplify decisions, and make room for better work.

  • AI applications
  • Knowledge & search
  • AI integration
Explore AI solutions
02

Purpose-built software.

Your business isn’t off the shelf. Your software shouldn’t be either. Build a product that works the way you need it to.

  • Web applications
  • Internal tools
  • APIs & integrations
Talk custom software
03

Smarter workflows.

Less busywork. More momentum. Bring your tools together and turn repetitive processes into things that just happen.

  • Process automation
  • Connected systems
  • Product prototypes
Find your next shortcut

02 / SELECTED WORK

Products and client work.

Built, and delivered.

Our own product, client engagements,
and systems our founder has shipped.

LiveQ: real-time AI copilot for interviews, meetings and calls, on macOS and Windows
OUR PRODUCTAI DESKTOP APP

LiveQ

A real-time AI copilot for interviews, meetings, and sales calls. It follows the conversation and what’s on screen, then offers answers and talking points in a private overlay.

1,500+users since launch in 2025
< 1 send-to-end latency for live transcription and coaching
  • Founded and built solo, on Next.js, Electron, LiveKit, and Convex
  • Native desktop app for macOS and Windows
  • AI mock interviewer and a code panel for technical practice
  • Live translation across five languages
  • Automatic summaries with decisions and next steps

Next.js · Electron · LiveKit · Convex

Visit liveq.ai
Illustrative image: a patient speaking with an AI voice consult on a desktop screen
CLIENT PROJECTVOICE & CHAT AI

Telehealth consult agent

An AI consult agent for a US telehealth company that talks with patients by text and voice. It was accurate but too slow for voice, so we rebuilt the response path to be fast without loosening the safety checks.

  • Profiled every turn end to end and found most of the delay was between services, not in the model
  • Moved reasoning into one streamed model call inside the app, beside the product catalog and safety logic
  • Built interruptible real-time voice on LiveKit and OpenAI’s realtime model, with every health answer still coming from the checked text pipeline
  • Shipped behind a feature flag, so the existing agent kept running unchanged

OpenAI API (text + realtime voice) · LiveKit Agents · TypeScript · Next.js · React · Python · Cloudflare Workers + Durable Objects · Neon Postgres · Vercel · Vitest · Playwright · GitHub Actions

Discuss a similar project
Illustrative image: aerial farm imagery with problem areas highlighted by a heat-map overlay
CLIENT PROJECTMACHINE LEARNING

Agricultural imagery analysis

A computer-vision model the client’s own team can train to find problem areas in agricultural imagery, picking out recognizable patterns with a geographic aspect.

  • Kickoff session to define the deliverable and agree the approach
  • Supervised model built to be retrained on the client’s imagery
  • Hands-on training in machine learning and computer vision practice
  • Continued support for further development

Python · PyTorch · Computer vision

Discuss a similar project
Illustrative image: a document parser turning scanned paperwork into structured output
CLIENT PROJECTLLM ENGINEERING

Multimodal LLM extraction

Hands-on AI/ML engineering with an AI consulting team: tuning and validating multimodal LLM workflows that turn unstructured documents into structured, reliable data for client-facing services.

  • Structured extraction and prediction from unstructured documents
  • Evaluation pipelines with LLM-as-a-judge, error analysis, and iterative testing
  • Guardrails and security review for high-fidelity output
  • Integration into client-facing services and pipelines

Python · LangChain · LangGraph · Langfuse · Pydantic · SQL

Discuss a similar project

More projects

BUILT AT COMMUREVOICE AI

Healthcare voice agent

A production voice AI agent for an enterprise healthcare organization: streaming speech-to-text, LLM, and text-to-speech, with an evaluation harness for regression testing and full observability across live traffic.

Python · Streaming STT + LLM + TTS · Tracing, metrics, alerting

40%+ of inbound calls automated

BUILT AT BRELLIUMDOCUMENT AI

Clinical document audit

An LLM-based audit platform for clinical records, with a generalized inference pipeline on vLLM for structured extraction and evaluation pipelines from prompt versioning to workflow observability.

vLLM · PyTorch · Python

90% less manual implementation time

BUILT AT STONK TECHSEARCH & MEDIA

Video search pipeline

A video and audio processing pipeline on AWS, built for throughput, feeding a vector-database query system that lets users search across media.

AWS (S3, Textract, CloudFront, Lightsail) · Vector database · MySQL

55% better search relevance and performance

BUILT AT STONK TECHLLM AGENT

Financial chatbot

A conversational finance assistant built on LangGraph, GPT models, and Ray for scalable serving.

LangGraph · GPT models · Ray

30% higher user retention and engagement

RESEARCHVOICE ASSISTANT

Alexa Prize TaskBot

An LLM-powered, voice-first multimodal task assistant built with the Penn State NLP Lab and deployed on Echo devices. One of 10 teams selected worldwide for Amazon’s Alexa Prize TaskBot Challenge 2.

Llama · CLIP · Alexa Skills Kit · GPU Docker · AWS (ECS, EC2, Lambda)

50% lower latency under heavy load

BUILT AT PENN STATEDOCUMENT AI

SmartOCR

An end-to-end collection and extraction pipeline for high school transcripts, built with the Nittany AI Alliance on AWS Textract, S3, and spaCy NER.

AWS Textract · spaCy NER · S3

95%+ accuracy on courses, grades, and credits

BUILT AT J&JOCR

Document OCR integration

AWS, GCP, and Tesseract OCR services integrated into existing document workflows at Johnson & Johnson, with field evaluation of algorithm performance.

Tesseract · AWS · GCP

10% higher OCR accuracy

OPEN SOURCEMACOS APP

MindWhisper

A native hold-to-talk dictation app for macOS. It streams speech to text and inserts it at the cursor in any application, running local-first on the user’s own API key.

Native macOS · Streaming speech-to-text

MIT licensed · 2026

Projects tagged “Built at” were delivered by our founder as an engineer at that organization.

03 / THE POSSIBILITIES

A little glimpse of what’s possible.

Ideas, in action.

Two illustrative concepts.
Imagine what we could build for you.

CONCEPT EXPLORATION

Knowledge, connected.

An AI workspace for answers that live across your business.

AI APPLICATION
CONCEPT EXPLORATION

Less friction. More flow.

A calmer home for projects, processes, and everyday work.

CUSTOM SOFTWARE

Concept projects shown for illustration. Your challenge comes next.

04 / OUR APPROACH

Human first.
Always.

Behind every useful piece of technology is a real person with a real problem. That’s where we start.

At TailWave.ai, we bring curiosity, thoughtful design, and practical engineering to the same table. The goal is simple: make things that make a difference.

Portrait of Jianxiang Gao, founder of TailWave.ai
Jianxiang (Allen) GaoFounder, TailWave.ai
Clear thinking. Careful building.
01

First, we listen.

We get to know your idea, your challenges, and the people you’re building for. Good questions lead to better foundations.

02

Then, we make it real.

From an early prototype to working software. We build in focused steps, with room for feedback along the way.

03

And keep making it better.

Test the details, learn from real use, and refine what matters. Thoughtful software has room to grow.

05 / KIND WORDS

In their words.

“Allen combines AI strategy and execution at a high level. His depth of understanding the objectives and landscape at rapid speed is unmatched.”

Joshua Host
Joshua HostCEO, Thrivelab

“Allen took our multimodal extraction workflows from promising to dependable. He built the evaluation around them, dug into every failure, and kept iterating until the structured output was something we could put in front of clients.”

Lawrence Ngo
Lawrence NgoCEO, Ngo Consulting Partners

“Allen built us a computer-vision model that our own team can keep training, and he took the time to teach us how. We came away with a working tool and a much better understanding of the methods behind it.”

James Creason
James CreasonOwner, Basin Ag Imaging

IT STARTS WITH A CONVERSATION

What’s your
next big idea?

A rough sketch. A recurring problem. A “what if.”
We’d love to hear it.

Opens your email app with the brief ready to send.