Scott Williams

About Me

Hi, I'm Scott, a Forward Deployed Engineer at Tenex (5th Engineer), based in New York City. I lead client-facing AI workflow and data engineering engagements, mainly focused on automating financial work and improving team efficiency. My recent independent work spans self-hosted operations, full-stack development, and GCP-based data pipelines applied to projects ranging from social media automation to personal finance research.

I learn and operate best by building. Recent projects include a data pipeline for automating arbitrage trades on Kalshi's prediction markets using order book and weather data, an AI-based video appraisal system, an MCP server for LLM debugging, tooling for rapidly deploying VPS-based full-stack applications, and LangGraph-based AI agents and workflows. These projects have deepened my expertise in cloud architecture, AI engineering, infrastructure as code, and end-to-end system design.

From June 2025 through January 2026, I worked as a Data Engineer at iCapital in New York City on the Data & Analytics team. I played a pivotal role in migrating mission-critical integrations from Airflow to Prefect and architected a Kafka-based streaming ingestion solution to bring real-time data from the platform into our data infrastructure.

I hold a Bachelor of Science in Computer Science and Applied Mathematics from the Colorado School of Mines. My academic background gave me a strong foundation in CS theory and its mathematical underpinnings, knowledge that continues to inform my work in the AI/ML space.

Outside of work, I'm a lifelong athlete. I've played hockey since age six and captained the rowing team at The Taft School. I also enjoy poker, hiking, running, and keeping up with AI research.

Thanks for visiting. Feel free to connect through the links above.

Work Experience

• Tenex, Forward Deployed Engineer (5th Engineer), New York, NY (January 2026 - Present):
  • Led 7 engagements across PE, CPG, and industrial companies, partnering with technical and nontechnical teams on internal AI, data engineering, and cloud infrastructure transformations.
  • Drafted and scoped SOWs and led technical sales calls for new prospects and contract expansions.
  • Architected and built headless agent systems to automate analysis and improve finance team decision-making:
    • Portfolio analytics: explained performance drivers and surfaced investment opportunities (Client ROI: 100 hours saved weekly across the team).
    • P&L reporting: explained forecast-to-actual variances from regional to SKU level (Client ROI: identified unprofitable products; informed leadership decisions).
    • Treasury forecasting: automated cash forecasting (Client ROI: 15 hours saved weekly; reduced overborrowing).
  • Developed LLM inference-heavy distributed systems supporting a consumer iOS app launch for approximately 20,000 users.
  • Delivered Salesforce solutions and data ingestion pipelines to automate client workflows.
  • Trained and adapted state-of-the-art Chinese open-source models using synthetic data and task-specific training and evaluation datasets.
  • Automated data-driven sales materials for CPG account managers, freeing time for relationships and new opportunities.
  • Collaborated with three selected engineers on Tenex’s first internal IP project.
  • Accumulated 750,000 impressions on technical X posts from July 2026 to October 2026 (@swill1ams).
• iCapital, Analyst - Data Platform Engineering Team, New York, NY (June 2025 - January 2026):
  • Used AI to enhance data quality and improve accessibility for technical and nontechnical stakeholders.
  • Migrated data pipeline orchestration tooling and refactored code and infrastructure for greater efficiency and scalability.
  • Partnered with fund finance teams to automate repetitive workflows, replace one-off Excel analyses with dashboards, and build robust data products.
• iCapital, Data Engineering Intern (Summer 2024):
  Researched Snowflake usage attribution and built DBT models supporting Tableau dashboards.
• Colorado School of Mines Career Center, Student Assistant (2022-2024):
  Coordinated Career Day and resolved technical issues for students and faculty.
• Coolipy, Open Source Contributor (February 2025 - Present):
  Collaborated with maintainers to identify and resolve bugs, contributing merged fixes to the Python client library for Coolify.

Awards

• Google Gemini NYC Hackathon 3rd Place ($20k Gemini Credits) (November 2025):
  Built an LLM debugging layer for coding agents. BugPoint is an LLM-native debugging system that provides AI coding agents with structured visibility into code execution through control-flow graphs and runtime traces, enabling agents to reason about complex control flow within tight context windows.
  Demo, GitHub Repo.
• Meta Llama 4 NYC Hackathon 2nd Place ($5k Prize) (June 2025):
  Competed in a 24 hour contest against hundreds of talented developers to build AthletIQ, a personalized AI coach which gives athletes feedback based on video footage from any sport. Achieved 2nd place overall in the contest.
  Demo Video, Hackathon Presentation, LinkedIn Post, GitHub Repo.

Projects

• SaaS Factory: Developed a CLI tool that automates VPS provisioning and configuration for rapid SaaS deployment. The tool spins up Linode instances (with plans to extend to additional providers like Hostinger and AWS EC2), automatically installs Coolify for VPS management, and configures open source products including N8N, Supabase, and Prefect. Additionally supports deployment of fully hosted Next.js starter applications with GitHub webhook integration for seamless CI/CD workflows, enabling developers to launch production-ready SaaS products in minutes with less than 5 commands and limited UI interaction. Currently documenting the tool to make it more user-friendly and accessible to developers of all experience levels.
• CaveatBot: Built a VS code extension in under 36 hours for a hackathon. The extension records all code changes, commands and their output, screenshots, and notes and adds them to a knowledgebase, allowing fast and efficient querying to recall configurations made to a project using Groq for fast inference and Pinecone's vector DB.
• Model Fine-Tuning & Synthetic Data Generation Micro-Project : Created a pipeline to filter and label reddit content for finetuning jobs using small open source models. This was ultimately used to fine tune the 1 billion parameter gemma 3 model to run locally for easy json extraction. Link to model: Hugging Face Model and link to dataset: Hugging Face Dataset.
• News Video Generator: This project automates the creation of short-form videos for TikTok and Instagram based on a topic prompt. It leverages web scraping to source recent news articles, utilizes semantic search and GPT-4-powered script generation, incorporates a custom text-to-speech API for narration, automates image sourcing, and integrates dynamic "white noise" video. Using FFMPEG for fast and automated video processing, the gathered assets are seamlessly organized and rendered into engaging, high-retention content optimized for social media platforms. See sample posts on Instagram: World News Daily & Tech News Daily.
• Personal Portfolio Builder: Created a chatbot to quickly generate and edit professional-quality portfolio websites.
• Clue Game Project: Recreated the game of Clue using Java with Test Driven Development and collaborative coding.

Skills

Education

• Colorado School of Mines, Golden, CO (Graduated May 2025)
  Bachelor of Science in Computer Science and Applied Mathematics
  Clubs and Involvement: Sports Analytics Club, Machine Learning Club, Ice Hockey
• The Taft School, Watertown, CT (Graduated May 2021)
  Clubs and Involvement: Varsity Crew Team (Captain), JV Ice Hockey, Investment Club, Robotics Club
• The Buckley School, New York, NY (Graduated May 2016)
  Clubs and Involvement: Tech Committee, Science Olympiad, Ice Hockey, Lacrosse
• Training the Street (Spring 2023)
  Excel modeling, basic accounting, asset valuation, cash flow analysis

Articles

• 
  The Rise of The Smoke Test

  Why coding agents reach for smoke tests, what those checks prove, and how to keep repositories maintainable as requirements change.

  Latest article · 9 min read · Read on X

• 
  Weaponizing Qwen 3.8: How Abliteration Removes an Open Model’s Security Guardrails

  An exploration of abliteration, model safety, and the tradeoffs of removing refusal behavior from open models.

  12 min read · Read on X · Read on LinkedIn

• 
  Kimi K3, explained: how forgetting made the biggest open model possible

  A look at Kimi K3’s memory and attention architecture, and the tradeoffs behind long-context efficiency.

  15 min read · Read on X