Shawn Yap
Zheng Yi

Solution Architect, AI Systems
BeyondEdge · Singapore

Client work is described, not named. The numbers come from shipped systems and finished research. Reach me at shawnyzy1@gmail.com. Recruiting? My resume is a one-page PDF.

What drives me is the intersection of language and intelligence, both artificial and human. Understanding how people encode meaning across languages shapes how I build systems that have to do the same.

Results

40% → 5%
Word error rate after Whisper LoRA fine-tuneFCS Solutions
30%
EER improvement in audio deepfake detectionHTX
+13%
Reply fidelity over base after LoRA fine-tuneBeyondEdge
200+
Intent classes with calibrated confidence thresholdsFCS Solutions
0.847 / 0.781
Accuracy and F1, BERT multi-label POFMA actor classificationNTU final-year project
32 vs 11
Trades rejected by the LLM risk reviewer, unbounded vs cappedSurvivor
676
Backtest sessions verifying no look-ahead biasSurvivor
5k+
Point-in-time forecasts scored with Brier calibrationSurvivor
10%
Whisper transcription accuracy gained on noisy audioHTX
1,300+
Social media comments classified for sentiment with BERTweetNTU final-year project
< 3 months
First commit to production on a full-stack operations platformWorkVer.io

Experience

  1. BeyondEdge

    Solution Architect, AI Systems

    Jul 2026 – Present · Singapore · Current

    Property AI Assistant

    Market-leading property agency

    • Led the build of the consumer-facing assistant, architecting its orchestration pipeline across 9 intents and co-developing backend API endpoints, with scope gating, conversational state management, and memory-aware query rewriting.
    • Engineered tiered retrieval across 4 Qdrant collections, falling back to partial to semantic matches, with geocoded proximity filters, user-confirmed search slots, and API-sourced numeric facts for grounded property recommendations.
    • Fine-tuned its Qwen3.5-9B model with LoRA on ~1k client-curated conversations, lifting reply fidelity 13% over base.

    Enterprise Agentic AI Platform

    Arm of a major global investment group · In production

    • Led the build and production deployment of a multi-tenant LLM agent platform with LiteLLM model routing, per-team token budgets, policy-gated MCP tools, and data-classification controls on GKE.
    • Prototyped a WhatsApp second-brain agent that distils free-text feedback into persistent memory over a 560-page LLM-wiki, productionising it into configurable department-level agents non-technical staff can tune without code.
    • Grounded agent briefs with per-claim citations to archived sources, prompt guards against forged citations, and watchdog alerts replacing silent run failures.

  2. WorkVer.io

    Co-Founder

    May 2026 – Present · Singapore · Current

    • Led the architecture and development of a full-stack operations platform covering jobs, quotations, invoicing, and payments, on Next.js, TypeScript, Drizzle ORM, Turso, and Vercel.
    • Built Telegram-based AI agents that query business data, execute workflows, and handle day-to-day admin through natural language.
    • Engineered English and Mandarin voice pipelines using Whisper STT and local TTS for hands-free agent use.
    • Built persistent agent memory through an LLM-maintained knowledge wiki of customers, sites, and past decisions.
    • Designed secure agent execution with sandboxing, RBAC, least-privilege access, and audit logging.
    • Reached production within 3 months of the first commit, using Claude Code and multi-agent coding workflows.

  3. FCS Solutions

    AI Engineer

    Jul 2025 – Jul 2026 · Singapore

    • Productionised an ASR–LLM–TTS pipeline with tool routing, schema validation, confidence gating, fallbacks, and telemetry.
    • Fine-tuned Whisper with LoRA, reducing WER from 40% to 5% and improving domain-specific intent recognition.
    • Built and fine-tuned a 200+ class intent-classification system from production logs, with calibrated confidence thresholds for production routing.
    • Built schema-aware NL-to-SQL agents over AWS Athena with constrained decoding, execution-based query validation, retries, and unsafe-query blocking.
    • Added agent reflection loops that use runtime failures and error analysis to improve SQL generation and schema grounding.
    • Owned the ML lifecycle from dataset curation and fine-tuning to benchmarking, deployment, and production monitoring.

  4. Home Team Science and Technology Agency (HTX)

    Data Scientist & Machine Learning Engineer Intern

    Oct 2024 – Jun 2025 · Singapore

    • Led R&D on audio deepfake detection for Singaporean English, using speech embeddings, prosodic features, and TTS-curated datasets, improving EER by 30%.
    • Built reproducible, seed-fixed benchmarks, evaluating via EER, ROC-AUC, precision-recall, and failure-mode analysis.
    • Designed denoising and augmentation pipelines, sweeping EER across SNR levels to characterise low-SNR robustness.
    • Developed a background noise reduction algorithm that improved Whisper transcription accuracy on noisy audio by 10%.
    • Prototyped LangChain LLM and RAG workflows for summarisation and contextual retrieval over internal documents, with an LCEL memory module for conversation continuity.

    Earlier internship: Data Scientist

    Dec 2023 – Mar 2024

    • Evaluated LLMs for a text processing tool through prompt engineering, hyperparameter tuning, and benchmarking.
    • Debugged a web-based WER calculator so transcripts in Latin and non-Latin scripts score correctly.

  5. Infineon Technologies

    Data Analyst and Automation Engineer Intern

    Jun 2023 – Aug 2023 · Singapore

    • Led development of Python data analytics GUI applications, using pandas and Matplotlib to analyse and visualise hardware testing data.
    • Proposed and prototyped a web application with React, Flask, and MongoDB as a central hub for processing data batches.
    • Road-tested the new applications and integrated them into the existing workflow to improve output volume and processing time.

Projects

  1. Survivor

    Project: evolutionary multi-agent trading system

    Jun 2026 – Present · Python, PostgreSQL, Docker, LLMs

    • Built a system that breeds, tests, and retires trading strategies by genetic search over 16 families and 8 regimes.
    • Scored 5k+ point-in-time forecasts against realised outcomes with Brier calibration, demoting overconfident strategies.
    • Capped the LLM reviewer to raising risk only; unbounded, it rejected all 32 trades (vs. 11) and halted trading.
    • Ran self-hosted Qwen inference under schema-constrained decoding with Pydantic validation of every model output.
    • Eliminated look-ahead bias structurally so backtests cannot access future data, verified over 676 backtest sessions.

  2. POFMA actor classification

    Final-year project: transformer classification and sentiment analysis

    Jan 2024 – Dec 2024 · PyTorch, BERT

    • Fine-tuned BERT for multi-label POFMA actor classification over 512-token chunks, at 0.847 accuracy and 0.781 F1.
    • Built scraping and PDF pipelines with contextual feature engineering, plus LLM sentiment analysis of public reaction.
    • Built custom Reddit and Facebook scrapers and classified 1,300+ comments with BERTweet, finding about 42% negative and 50% neutral.
    • Combined the NLP analysis with public policy research into regulatory enforcement and public discourse.

    Source

Hackathons

  • TechfestJan 2023Telegram bot in Python that gamifies reporting HDB car park faults, so authorities can track and manage them.
  • MLDA@EEE Deep Learning WeekJul 2022Computer vision model on a live video feed to identify the presence and activity of elderly people living alone.
  • DSTA BrainHackDec 2021Team of 4 training a pose detection model with YOLOv3 and an NLP model for emotion in speech.

Public repositories

  • quantwachstum

    Self-hosted equity research triage: transparent scoring, scenario ranges, Telegram delivery.

  • pofma-sentiment-analysis

    Actor classification and sentiment analysis over every POFMA notice since 2019.

  • htx-dl-denoise

    Noise reduction with DCCRNet and traditional audio engineering.

Skills

AI Engineering
Python, Async Python, FastAPI, REST APIs, SQLAlchemy, PostgreSQL, Docker, Git, AWS, GCP, Azure, Pytest, Automated Testing, Model Deployment, Claude Code
Data Science
pandas, NumPy, SQL, Feature Engineering, Statistical Analysis, Experiment Design, Backtesting, Point-in-Time Evaluation
Machine Learning
PyTorch, Hugging Face Transformers, TensorFlow, LoRA/PEFT, Fine-tuning, BERT, Qwen, Whisper, NLP, ASR, Embeddings, Dataset Curation
Agentic AI & LLM Systems
Tool Calling, Tool Routing, MCP, RAG, Hybrid Retrieval, Qdrant, Structured Outputs, Constrained Decoding, NL-to-SQL, Prompt Engineering, LangChain, OpenClaw
LLM Reliability & Evaluation
Guardrails, Confidence Gating, Schema/Execution Validation, Retry Logic, Calibration, Threshold Tuning, WER, EER, F1, ROC-AUC, Brier Score, Error Analysis, Benchmarking
Full-Stack & Platform
Next.js, TypeScript, React, Flask, MongoDB, Drizzle ORM, Turso, Vercel, AWS Athena, RBAC, Sandboxing, Audit Logging

Education

Nanyang Technological University
Bachelor of Computing (Hons), Computer Science, Minor in Modern LanguagesFootball Captain, Hall of Residence 12, for three years. Assistant Chief Group Leader, College of Computing and Data Science orientation.Singapore · Aug 2021 – May 2025
University of Stuttgart
Semester exchange and German immersion programme, Computer ScienceModules: Machine Learning, Virtual and Augmented Reality, Computer Security.Germany · Mar 2024 – Aug 2024

Certifications

  • DeepLearning.AI TensorFlow Developer Specialization
  • Google Advanced Data Analytics Certificate
  • Introduction to Machine Learning in Production, DeepLearning.AI
  • Unsupervised Learning, Recommenders, Reinforcement Learning, DeepLearning.AI
  • Python for Data Science, AI & Development, IBM

Languages

  • EnglishNative
  • Mandarin ChineseNative
  • CantoneseWorking proficiency
  • GermanCEFR B1, learning
  • MalayLimited working
  • SpanishElementary, learning
shawnyzy1@gmail.com