# Lam Pham (Pham Thanh Lam)

> Lam Pham (Phạm Thanh Lâm, also written "Laam") is Chief Data Science & AI Officer at Cake by VPBank, a digital bank in Vietnam named Best Digital Bank in Vietnam 2026 by Euromoney. He leads the DSAI team that builds the models behind credit decisions and revenue forecasting, and runs a fleet of 150+ voice agents that have handled 25M+ real customer calls. Based in Ho Chi Minh City, Vietnam. Personal site: https://www.lampt.org

Motto: "Small tools, large systems. AI that handles real money."

## Key facts

- Current role (2024–present): Chief Data Science & AI Officer (CDAO), Cake by VPBank. Cake has ~8M customers, reached EBITDA break-even in Q3 2024 (3.5 years after launch), and grew revenue per customer from $5 (2023) to $25 (2025).
- Current work: moving Cake from digital bank to "Next Gen AI Bank": causal inference for credit risk, multi-agent orchestration, voice-agent fleet management and evaluation (SNR monitoring, golden QA eval sets), and an intel-brief pipeline (PDF → markdown → SQL) that cut the cost of reading a Vietnamese financial report from $8 to $0.80.
- 2019–2023: Co-founder of TrueID and Head of AI Products at VNG Corporation (Vietnam's first tech unicorn). Digital identity and eKYC for financial services; $1M+ ARR AI business, 10M+ users (B2B2C). eKYC liveness confirmed under iBeta ISO 30107-3 PAD; face recognition benchmarked in NIST FRVT 1:1.
- 2015–2018: Lead Data Scientist at Viettel (telco), Sentifi AG (financial sentiment, Switzerland) and VNG (gaming, social).
- 2012–2015: Deputy Head of Mobile at GNT Inc (Mobion Music, #1 on the Japan App Store), FPT Software, co-founder of Saigonapps.
- Education: MSc Computer Science, Saint Petersburg State University, Russia (2001–2008); MSc Business Information Systems, Vietnamese-German University (2010–2012).

## Achievements

- 1st place, WSDM Cup 2019 (ACM WSDM), fake news detection (ByteDance task), 700+ teams, modified BERT: https://github.com/lampts/wsdm19cup
- Kaggle Master (handle: saigonapps). CIKM 2017 (4th), CIKM 2018 (8th), AICovid Vietnam 2021 (2nd).
- Keynote, IFC / World Bank FIDN 2020: "Implementing Digital Identity Platforms for a Digital Economy".
- Speaker: PyCon Singapore 2016 (music2vec, song arithmetic from million playlists), TopDev 2019 (Productionizing AI), VietnamWorks Tech Expo 2016.
- Open source: bert4vn and albert4vn, among the first openly released Vietnamese language models (2019). VietAI open lectures (2017).
- Press: TheLEADER, "Sự trỗi dậy của Cake: Từ Việt Nam đến hình mẫu ngân hàng số khu vực" (Sep 2026): https://theleader.vn/su-troi-day-cua-cake-tu-viet-nam-den-hinh-mau-ngan-hang-so-khu-vuc-d47783.html

## How he builds

- Unix philosophy: small composable tools that multiply force.
- Slope > intercept: hire and build for learning rate and compounding.
- Ship > paper: a model in production beats a model in a notebook.
- Hard to vary: if an element is decorative, remove it.

## Profiles

- [Homepage](https://www.lampt.org/): bio, career, credentials (canonical source for facts above)
- [GitHub](https://github.com/lampts)
- [LinkedIn](https://www.linkedin.com/in/lam-pham-thanh-1428602b)
- [Kaggle](https://www.kaggle.com/saigonapps)
- Email: laampt@gmail.com

<!-- generated:start -->
## Tools (65 standalone, single-file HTML apps; index: https://www.lampt.org/tools/)

### ai
- [Token Rate — How Fast Is N tok/s?](https://www.lampt.org/tools/tok.html): How fast is N tokens per second? Stream any web page at a chosen LLM token rate, or compare two rates side by side, to feel generation speed. (2026-09-21)
- [The KV Cache — Interactive Monograph](https://www.lampt.org/tools/kvcache.html): An interactive monograph on the KV cache: why naive decoding is O(n³), what Q, K and V do, one decode step, the size of the cache, the causal mask, memory bandwidth, and prompt caching. With inline demos and a closing assessment. (2026-09-20)
- [Gemini Live — Voice Chat](https://www.lampt.org/tools/live.html): Talk to Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking with live audio, transcripts, and optional document context. (2026-09-19)
- [Probability Calibration — Interactive Monograph](https://www.lampt.org/tools/calib.html): Probability calibration, an interactive monograph: reliability diagrams, Platt and isotonic scaling and proper scoring rules, run live with Python in the browser. (2026-09-19)
- [Gemini Flash TTS — Cost & Lyric Sync](https://www.lampt.org/tools/tts.html): Single & multi-speaker text-to-speech via Gemini Flash TTS, with live token counting and cost estimation. (2026-06-21)
- [microGPT – Karpathy](https://www.lampt.org/tools/gpt.html): Karpathy's microGPT, annotated: walk through the 243-line, dependency-free Python that trains and runs a GPT, from tokenizer and autograd to Adam. (2026-02-12)
- [AI GTM Playbook](https://www.lampt.org/tools/gtm.html): AI go-to-market playbook: an interactive Bullseye framework to brainstorm 25 traction channels, test a few and focus on the one that works. (2026-01-10)
- [AI Market Trends 2025-2030](https://www.lampt.org/tools/ai30.html): AI market trends 2025–2030: an interactive strategy board of enterprise AI trends (agentic, sovereign, physical AI) and the pilot-to-scale gap. (2025-11-29)
- [AI Skills Gap Heatmap](https://www.lampt.org/tools/skill25.html): AI skills gap heatmap, 2015–2025: animated demand intensity by skill and year, based on WEF Future of Jobs and LinkedIn skills data. (2025-11-29)
- [LLM Comparison](https://www.lampt.org/tools/llm.html): Compare LLMs side by side: send one prompt and system message to OpenAI, Anthropic and Gemini models and read their answers together. (2025-11-27)
- [LLM Council](https://www.lampt.org/tools/council.html): LLM Council: several LLMs answer your query, review and rank each other's answers, then a chairman model synthesizes a final response. (2025-11-27)
- [PictureMe – AI Photo](https://www.lampt.org/pictureme.html): PictureMe: transform your photo into themed portraits with Gemini AI. Use your camera or upload, pick a theme or write a custom style. (2025-10-08)
- [HeadshotPro AI](https://www.lampt.org/headshot.html): HeadshotPro AI: turn any photo into a professional headshot with Gemini. Upload, choose a style (corporate, creative and more), download. (2025-10-04)
- [Direct Photo Analysis Tool](https://www.lampt.org/photo.html): Photo analysis tool: paste a photo URL and get professional photography feedback on composition, light and technique from Claude. (2025-08-16)
- [Ao Dai Benchmark – AI Rankings](https://www.lampt.org/aodaibench.html): Ao Dai Benchmark: which AI model draws the best SVG of a Vietnamese girl in ao dai with a conical hat and lotus flowers? Pairwise rankings of 14 models. (2025-06-07)
- [Loss Landscapes – 3D Viewer](https://www.lampt.org/loss.html): Explore 3D loss landscapes of GPT-2, LLaMA-3.2-1B, DeepSeek-R1-Distill-8B, Gemma-2, Qwen2 and Qwen2.5. Rotate, pan and zoom each surface. (2025-05-24)
- [Interactive Meta-Prompt Configurator](https://www.lampt.org/prompt.html): Interactive meta-prompt configurator: assemble a structured system prompt from configurable sections and get word, character and line statistics. (2025-05-08)
- [LeCun Net Digit Recognizer](https://www.lampt.org/lecun.html): Draw a digit and watch a LeCun-style convolutional network classify it in the browser via ONNX, with H1/H2 feature maps and softmax probabilities. (2025-05-04)

### dev
- [minidraw — Shared Canvas](https://www.lampt.org/tools/draw.html): minidraw: a tiny shared canvas. Draw with pen, shapes and text, then share the whole board inside one URL. No server, no account. (2026-06-14)
- [Python ML Stack – Deps](https://www.lampt.org/tools/xz.html): Python ML stack dependency graph: explore how frameworks, core libraries, compute and system packages depend on each other, with xz-style risk flags. (2026-02-27)
- [Jina Reader Proxy](https://www.lampt.org/tools/proxy.html): Jina Reader proxy: extract any web page as markdown, HTML or text, then optionally summarize it with Claude Haiku. (2025-12-23)
- [PDF to Gist](https://www.lampt.org/tools/p2g.html): PDF to Gist: extract PDF pages as compressed images in the browser and publish them to a GitHub Gist with your token. (2025-12-03)
- [nanotele](https://www.lampt.org/tools/tele.html): Cut through noise. Extract signal from Telegram group chats. (2025-12-02)
- [HBR Year in Tech 2026](https://www.lampt.org/tools/hbr.html): HBR Year in Tech 2026, analyzed: explore 570+ insights from 10 articles as claims, metrics and actions, filterable and in a timeline view. (2025-11-29)
- [PyConsole TUI](https://www.lampt.org/tools/code.html): PyConsole: a terminal-style Python console in the browser with NumPy and pandas, powered by Pyodide. Shift+Enter to run. (2025-11-25)
- [Crypto Signing & Verification](https://www.lampt.org/crypto.html): Interactive playground for cryptographic signatures: pick public parameters and a secret key, then sign and verify messages step by step. (2025-10-12)
- [QR Code to Go Board Converter](https://www.lampt.org/qr.html): Turn any text or URL into a QR code drawn as black and white Go stones on a Go board. Inspired by Veritasium. (2025-06-12)
- [Reservoir-Sampling Log Rate Limiter](https://www.lampt.org/reservoir.html): Reservoir sampling explained as a log rate limiter: an animated visualization of keeping at most K random logs per second from a stream. (2025-05-22)
- [WebAssembly IPython](https://www.lampt.org/ipython.html): Run Python with NumPy in the browser. A minimal IPython-style console powered by Pyodide and WebAssembly, no install needed. (2025-05-01)

### finance
- [Tục Ngữ Tài Chính](https://www.lampt.org/tools/sa.html): Tục ngữ tài chính: 7 Vietnamese proverbs as timeless financial principles, each proven with an interactive simulation (in Vietnamese). (2026-03-05)
- [Dodge-meter – Earnings Call](https://www.lampt.org/tools/ometer.html): Dodge-meter: score earnings-call Q&A for evasion. 31 linguistic pattern categories rate question sharpness and answer evasiveness, with optional LLM review. (2026-01-18)
- [TVM Calculator](https://www.lampt.org/tvm_calculator.html): Time value of money (TVM) calculator: solve for PV, PMT, FV, rate or periods with monthly, quarterly, semi-annual or annual compounding. (2025-04-27)

### fun
- [In My Life – Solo Guitar](https://www.lampt.org/tools/solo.html): In My Life solo guitar masterclass: a 12-module path with audio, an interactive fretboard and a lick library (explained in Vietnamese). (2026-01-25)
- [Lucky Draw Simulator](https://www.lampt.org/tools/luck.html): Simulate any probability event. See the true cost. (2026-01-10)
- [Guitar Lead Trainer](https://www.lampt.org/tools/guitar.html): Guitar lead trainer: load an MP3 or WAV, change speed and pitch independently, loop an A–B section and practice with a tap-tempo metronome. (2026-01-02)
- [Poker Trainer](https://www.lampt.org/tools/poker.html): Texas Hold'em poker trainer: play hands against villain profiles with a step-by-step decision framework, M-ratio zones and expected value. (2025-12-30)
- [Dice Roll Simulator](https://www.lampt.org/tools/baby.html): Dice roll simulator for the true cost of a rare event: roll four dice until a ~1/144 outcome appears at $5 per roll, like a blind-box unboxing. (2025-11-18)
- [Progressive Metal & Rock](https://www.lampt.org/rock.html): Progressive metal and rock as a metro line, 1965–2012: play landmark tracks from bands that broke the line between complexity and brutality. (2025-10-09)
- [Classical Music Metro Line](https://www.lampt.org/classic.html): Classical music as a metro line, 1678–1893: click any station to play a famous piece. Circle size reflects popularity adjusted for age. (2025-10-06)
- [Paws & Coffee](https://www.lampt.org/paws.html): Paws & Coffee: a demo website for a dog café in Saigon (Vietnamese), with menu, dogs and contact sections. (2025-06-26)
- [Bird Stories – by AI](https://www.lampt.org/story.html): Bird Stories: an illustrated collection of short fables about birds, written with AI. (2025-06-02)
- [Mini Spotify – Offline Player](https://www.lampt.org/mp3.html): Mini Spotify: an offline music player in one HTML file. Upload MP3 or WAV files, build a playlist and play them locally in the browser. (2025-05-25)
- [The Book of Answers](https://www.lampt.org/tboa.html): The Book of Answers: ask a serious question, meditate, click the crystal ball and receive an answer. A playful digital oracle. (2025-04-27)

### learning
- [The Petersburg Line — Science Metro](https://www.lampt.org/tools/pantheon.html): The Petersburg Line: a time-travel metro of science. 21 stations across 300 years, each a scientist who worked on Vasilievsky Island, St. Petersburg. (2026-09-13)
- [Drill Forge — Pronunciation Drills](https://www.lampt.org/tools/drill_forge.html): Drill Forge: generate a 120-word English pronunciation drill covering all 44 sounds, set in a world you pick (a band, writer or fairy tale), in RP, American and other accents. (2026-08-23)
- [Experiment Valley – Competition](https://www.lampt.org/tools/abc.html): Experiment Valley Competition - Test your A/B testing skills against AI opponents. (2026-02-12)
- [Experiment Valley – Tutorial](https://www.lampt.org/tools/ab.html): Experiment Valley Tutorial - Learn A/B testing through interactive lessons with Python code. (2026-02-11)
- [ML/DS Quiz Platform](https://www.lampt.org/tools/quiz.html): ML/DS quiz platform: generate machine learning and data science assessments, share them via URL and track results. No server; data lives in the URL. (2026-02-04)
- [AI English Teacher](https://www.lampt.org/tools/teach.html): AI English teacher: practice spoken English in real time with a voice tutor built on the OpenAI Realtime API. Choose voice and teaching style. (2025-12-11)
- [NanoFluency: Native Echo](https://www.lampt.org/tools/speak.html): NanoFluency Native Echo: practice English pronunciation by matching the timing and rhythm of AI-generated native speech, word by word. (2025-11-23)
- [Chinese Learner – HSK 1-2](https://www.lampt.org/tools/hellochinese.html): Learn Chinese for HSK 1–2: a guided lesson roadmap with progress tracking for beginner Mandarin vocabulary and phrases. (2025-11-16)
- [Flashcard App](https://www.lampt.org/card.html): Flashcard app with flip, keyboard navigation, search, card creation and progress statistics. Ships with Chinese–English example cards. (2025-05-23)

### viz
- [Marigold Bay — Isometric City](https://www.lampt.org/tools/city.html): Marigold Bay: a living isometric city simulation with citizens, vehicles and a day clock. Pan, zoom and change the simulation speed. (2026-06-14)
- [Animated Word Cloud](https://www.lampt.org/tools/wc.html): Animated word cloud: paste text and watch the Archimedean spiral placement algorithm build the cloud step by step. Export as PNG. (2026-03-01)
- [KB Visualization](https://www.lampt.org/tools/kbvis.html): Knowledge-base visualization: embed entries with TF-IDF, project them with UMAP and explore clusters with click and lasso selection. (2026-01-15)
- [Forest Fire Model](https://www.lampt.org/tools/fire.html): Self-organized criticality simulation - Drossel-Schwabl forest fire model (2025-12-27)
- [Collatz Conjecture (3N+1)](https://www.lampt.org/collatz.html): Collatz conjecture (3N+1) visualizer: animated trajectories, step and peak statistics, and a Benford's law check of leading digits. (2025-10-14)
- [Interactive SVG Comparison](https://www.lampt.org/svg.html): Compare two SVGs side by side: paste code, drop files or load by URL (?left= / ?right=) into Before and After panels. (2025-06-07)
- [Solar System Explorer](https://www.lampt.org/solar.html): Interactive solar system explorer: animated planets orbiting the Sun; click any body for facts and adjust the animation speed. (2025-05-23)
- [Waveform Visualizer](https://www.lampt.org/wave.html): Waveform visualizer: full-screen animated, glowing sine waves layered with different frequencies, speeds and amplitudes on canvas. (2025-05-22)
- [Galton Board Simulation](https://www.lampt.org/gemini_galton.html): Galton board simulation: drop balls through rows of pegs at an adjustable rate and watch the binomial distribution build into a bell curve. (2025-04-25)

### writing
- [LexMark](https://www.lampt.org/tools/le.html): LexMark: read Vietnamese legal and banking-compliance documents with AI term explanations for lawyers, risk officers and auditors. (2026-03-06)
- [Reader](https://www.lampt.org/tools/reader.html): EPUB reader in the browser: a library, chapter navigation, dark mode and layout settings, plus AI dictionary lookups via Claude, GPT or Gemini. (2026-02-20)
- [Inkflow – Focus Reader](https://www.lampt.org/tools/tao.html): Inkflow: a distraction-free focus reader and writer with highlighting, underlining and Ask AI (fix, rewrite, translate, explain). Share via URL or QR. (2025-11-22)
- [Annotated Presentation Creator](https://www.lampt.org/annotate.html): Annotated presentation creator: load slide images, OCR them for alt text, and generate an annotated write-up of a talk. (2025-05-16)

## Notes (TIL explainers)

- [Bayesian Thinking | Interactive Tutorial](https://www.lampt.org/til/bayesian.html): Learn the most powerful framework for reasoning under uncertainty.
- [How to Deploy AI Agents in Production: The Falconer's Framework](https://www.lampt.org/til/falcon.html): A practical framework for deploying AI agents at scale, based on running 150+ voice agents at Vietnam's largest digital bank. Context engineering, verification patterns, and failure modes.
- [KB Agent System Design — arxiv-sanity Style](https://www.lampt.org/til/kba.html): Zero-infrastructure KB Agent for Banking Voice Agents
- [KV Caching: Why Cached Tokens Are 10x Cheaper](https://www.lampt.org/til/kv.html): Interactive tutorial on KV caching in LLMs
- [Logprobs: Teaching LLMs What They Don't Know](https://www.lampt.org/til/logprob.html): An interactive tutorial on using log probabilities for uncertainty quantification in LLM applications.
- [Interactive Network Cascade](https://www.lampt.org/til/network.html)
- [Signal vs. Noise](https://www.lampt.org/til/overthink.html)
- [The Magic Number | Riffle Shuffle Tutorial](https://www.lampt.org/til/riffle.html): Interactive tutorial on why 7 riffle shuffles randomize a deck - based on Diaconis-Bayer theorem
- [Agent Skills Mastery](https://www.lampt.org/til/skill.html)
- [SNR Tutorial - Signal-to-Noise Ratio](https://www.lampt.org/til/snr.html): Learn Signal-to-Noise Ratio in 10 minutes. Interactive tutorial.
<!-- generated:end -->

## Optional

- [Cake by VPBank](https://cake.vn): the bank where he works
- [music2vec talk, PyCon SG 2016](https://pyvideo.org/pycon-singapore-2016/music2vec-a-smart-music-recommendation-and-discovery-pyconsg-2016.html)