Neo / AI & LANGUAGE MODELS
NEO / AI RESEARCHER

Language.
Intelligence.
Imagination.

Hi, I’m Qiang Gao.
AI Researcher at ByteDance.
Studying intelligence. Leaving room for wonder.

A little about me ↗
FIG. 01 / LATENT SPACE3D FIELD
+
+
THOUGHT, TAKING SHAPE.
A RESEARCH MIND. A WANDERING EYE.SCROLL TO EXPLORE ↓
01 / ABOUTFROM RESEARCH TO PRACTICE

Curiosity is
the throughline.

I’m Qiang Gao — call me Neo — an AI Researcher at ByteDance since September 2026. My work connects language model research with real-world search and relevance systems.

I earned my master’s degree at Wuhan University, advised by Prof. Fei Li, studying information extraction and retrieval-augmented generation. Previously, I worked on LLM-powered search at Alibaba Xianyu and interned at Tencent AI Lab, exploring mixture-of-experts models and conversational game NPCs.

A little rigor. A little reverie. Always curiosity.↗
02 / FOCUSAN ONGOING EXPLORATION

Working with
language & intelligence.

The field I work in.
The questions I keep coming back to.

[ 01 ]
Aa_

Language models

Mixture-of-experts architectures, supervised fine-tuning, preference optimization, and efficient inference.

LANGUAGE
[ 02 ]
ƒ(x)

Search & relevance

LLM-based evaluation, model distillation, query understanding, and retrieval-augmented generation.

RETRIEVAL
[ 03 ]
↗

Information extraction

Cross-document event extraction, discourse-aware representations, and faithful generation.

RESEARCH
03 / PUBLICATIONSSELECTED RESEARCH

Questions worth
looking into.

All publications on Google Scholar ↗
04 / JOURNEYTHE PATH SO FAR

A new chapter.
The same curiosity.

SEP 2026
— PRESENT
CURRENT

ByteDance

AI Researcher

AI & language models
JUN 2025
— 2026
INDUSTRY

Alibaba · Xianyu

LLM Researcher

Search relevance, evaluation, reranking, and conversational AI search.
FEB 2024
— DEC 2024
INTERNSHIP

Tencent AI Lab

LLM Research Intern

Mentored by Jian Li. Warm-start MoE models and conversational game NPCs.
SEP 2022
— JUN 2025
MASTER’S DEGREE

Wuhan University

Natural Language Processing

Advisor: Fei Li
National Scholarship · Academic Excellence Scholarship
SEP 2018
— JUN 2022
BACHELOR’S DEGREE

East China University of Science and Technology

Energy and Power Engineering
Second major in Computer Science and Technology

Top 10% · National Encouragement Scholarship
05 / THE PLAYGROUNDART, ALGORITHMS & WHAT-IFS

Some ideas begin
with “what if?”

A sketchbook for the less obvious questions.
Small experiments. Unfinished possibilities.

01 / INTERACTIVE SKETCH

A word, becoming a world.

What if a thought could leave a visual fingerprint? Give this little drawing machine a phrase and watch it find a shape.

A procedural drawing, inspired by generative art. Runs entirely in your browser; no AI model or upload.

a quiet kind of intelligenceSKETCH 001
02 / CONCEPT NOTE

The museum of
misunderstandings.

An imagined exhibition where two models keep describing the same picture to each other. What disappears? What becomes a story?

AI × PERCEPTION × ART
03 / CONCEPT NOTE

A soundtrack
for a sentence.

Imagine translating a paragraph’s rhythm into sound: punctuation becomes silence, repetition becomes a motif, and ambiguity becomes harmony.

LANGUAGE × MUSIC
06 / OFF THE CLOCKA LIFE BEYOND THE LOSS CURVE

Room for
other kinds of attention.

Reading, hiking, tennis, cycling.
Different rhythms, the same curiosity.

01 / READING

Other minds,
other worlds.

Books offer another way to travel: through an idea, a life, or a question that stays open.

PAGES / PERSPECTIVES
02 / HIKING

A longer
view.

Time on a trail. A changing horizon. Space for thoughts to settle at their own pace.

TRAILS / OPEN AIR
03 / TENNIS

One point
at a time.

Rhythm, attention, and the small pleasure of meeting the ball just right.

MOVEMENT / FOCUS
04 / CYCLING

Take the
scenic route.

A different speed for seeing the world. New roads, a little wind, and room to wander.

ROADS / DISCOVERY
07 / SELECTED WORKRESEARCH IN PRACTICE

Ideas, out
in the world.

Selected industry projects,
collected by company.

Alibaba · Xianyu

2025 — 2026 / LLM RESEARCH

Language models for search, relevance, and discovery.

LLM-powered relevance evaluation

Built an automated evaluation pipeline combining supervised fine-tuning, difficult-example augmentation, retrieval of similar samples, and DPO preference alignment.

96% agreement with human labels8 → 3 days annotation turnaround

SFT · RAG · DPO · Automated evaluation

Efficient LLM search reranking

Distilled a 7B teacher into a compact 0.4B discriminative reranker based on Qwen2.5. Built the serving pipeline and optimized inference with caching and FP8 quantization to extend real-time relevance scoring to long-tail queries.

30.95% fewer relevance bad cases in the 8M+ query segment

Knowledge distillation · Qwen · Model serving

Conversational AI search

Developed a three-stage query understanding architecture: contextual intent understanding, query rewriting, and structured search-condition parsing. Improved relevance screening with an upgraded Qwen 7B model and retrieval-augmented knowledge.

70% → 95% usable query rewrites8% → 3% empty-result rate

Query understanding · Multi-turn context · RAG

Tencent AI Lab

2024 / RESEARCH INTERNSHIP

Mixture-of-experts models and characters that remember.

Warm-start MoE & conversational NPCs

Explored Yi-based 4-expert and 8-expert MoE models for game dialogue. Designed an NPC architecture combining event memory, knowledge retrieval, and response generation. Built a Llama 3-based MoE project with bilingual fine-tuning, YaRN context extension, and vLLM serving.

Explore llama3-8x8b-MoE ↗

Mixture of experts · Memory · YaRN · vLLM