Built, not
commissioned.
Seven tools, pipelines, and data products I built as core infrastructure for a global nonprofit accelerator managing 4,486 startups. When generative AI landed commercially, I stopped waiting for a data team and started building. The same pattern I recognized at One Degree Solar with LED costs in 2011, and at Formlabs with desktop SLA in 2017: a technology crossing a threshold.
The operating question was: what does AI actually unlock for a lean team?
By early 2024, MassChallenge had 4,486 alumni companies in a database with no reliable sector taxonomy, no way to search by pitch text, no survival model, and no operational tooling that wasn't living in spreadsheets or static decks. A data team request would have taken months. I built the foundation in weeks.
The classifier came first. That unlocked the Explorer. The Explorer unlocked the Dashboard. The Dashboard unlocked the LinkedIn data journalism series. Each tool was infrastructure for the next one. The order mattered.
Everything here was built solo. Python/pandas for data work. Anthropic API directly for classification and content generation. Netlify for static deploys and serverless functions. No frameworks that required a build pipeline. No IT tickets. That's the point: a non-traditional technical operator who can ship working tools rather than spec them for someone else.
Seven tools. Each one solving a real operational problem.
Crucible
Most accelerator programs evaluate companies the same way: spreadsheets, gut feel, and whoever has the most domain knowledge in the room. Crucible replaces that with something more rigorous. An HC&LS company submits an application; within minutes the system has run live web research on the founders and market, looked up their FDA regulatory pathway, pulled CMS reimbursement data, and scored them across four frameworks: IDEO human-centered design, MIT Disciplined Entrepreneurship, Clinical Science maturity, and a Revenue Bridge analysis. That used to take a program team days.
The architecture is three-pass, not a wrapper. Sonnet with web search runs the initial evaluation and captures every query it ran and what it found. Haiku critiques the output against four specific criteria: FDA pathway accuracy, reimbursement specificity, risk severity calibration, and score-recommendation alignment. If it flags a problem, Sonnet revises only the failed fields. The reasoning trace is visible in the UI, collapsed by default. Specialists, who only see applications the AI flagged for genuine domain uncertainty, see more of it. Their time is treated as the scarcest resource in the system.
The scoring rubric is built around one question: does this company's survival depend on a future binary event it can't control? A regulatory approval it hasn't gotten. A reimbursement code that doesn't exist yet. The top band (90–100) is reserved for companies that don't have that dependency. The demo cohort is 12 real HC&LS companies, all publicly raised Series A: Hippocratic AI, Tennr, Abridge, Rad AI, Eko Health, Proprio, Cala Health, Pomelo Care, Nourish, Avenda Health, Proscia, and Boundless Bio. Every evaluation runs on publicly available information only.
Field Notes
A private deal intelligence tool for early-stage VC and advisory work. Three capabilities, all connected through a persistent company record: an investment memo builder structured around the Barley/Roberts HBS framework (why now, why this, why these people, traction read, what would have to be true, risks, recommendation) with inline framework cards citing Andreessen on markets and Botha on "what would have to be true," plus AI-generated IC pitch and skeptical-partner stress test; a term sheet analyzer that translates every clause into plain language and tags it Market/Aggressive/Founder-friendly against Q1 2026 benchmarks (Aumni, Carta, NVCA), with a VC Lens and Founder Lens toggle and exit waterfall at four scenarios; and a fund profile and portfolio construction model showing investable capital, companies the math supports, and three return scenarios, with every term sheet showing a deal-vs-fund analysis: ownership achieved vs. target, percent of fund consumed, reserves needed, the multiple this single deal has to return the whole fund.
The integration is the point. Cap table tools (Carta, Pulley) and term sheet templates (NVCA, YC) exist separately. None connect memo writing to term sheet analysis to fund construction logic. The Berkeley VCEP framework is the through-line: the tool encodes what a VC partner with formal training actually thinks, with source attribution visible in the UI. Waterfall math handles the common case correctly and explicitly flags edge cases, like pari passu structures and participation caps, rather than producing confidently wrong numbers.
Built in extended conversation with Claude across a single working session. Three versions shipped: v1 (memos, term sheets, AI synthesis), v2 (lifecycle events, multi-round stacked waterfall, shareable links with viewer tracking), v3 (fund profile, portfolio construction, deal-vs-fund integration, in-app admin invite system). An early scoring prototype was scrapped for being too accelerator-oriented and rebuilt from investor-standard memo structure. About 4,500 lines of TypeScript/TSX.
Accelerate v2: Program Management Platform
MassChallenge runs 12+ programs across Boston, Texas, Israel, and Switzerland using seven separate systems. No single view answers "how is this cohort doing" without opening at least four tabs. AcceleratorApp's contract ends July 2026, right after HC Traction 2026 closes. That's the ship date. Accelerate v2 replaces AcceleratorApp, the legacy Accelerate platform, and the Airtable workflows that accumulated in the gaps between them.
A 6,957-line single-file interactive prototype: React via CDN, Babel in-browser, no backend, no build pipeline, realistic demo data seeded for HC Traction 2026. Every workflow in the PRD is click-through interactive. Seven user personas, each with a fully scoped role view. Three program types, each with its own module configuration: Traction (20–40 startups, 4 months, mentor-centric), Challenge (2–5 corporate partners each selecting 1–4 startups, 3 months), Custom (single partner, 1–10 startups, 6–8 weeks, designed to launch in under two weeks from contract).
Eight AI agents underneath, each producing a recommendation a human accepts, edits, or rejects. Screener triages incoming applications before they reach the PM queue. Allocator distributes judge load by expertise match. Synthesizer rolls up R1 and R2 scores and proposes a cut line. Matchmaker pairs mentors to startups with rationale that propagates to both parties' views. Coach pre-drafts biweekly progress reports from session activity so mentors edit rather than write from scratch. Curator scores applications against each partner's challenge statement in real time. Scout surfaces alumni-to-mentor pipeline candidates. Conductor orchestrates all seven and maintains the audit log. Every agent output is visually distinguished in MC Purple, the only use of that color in the design system. Total production agent cost: approximately $103/month. Platform ceiling: $500/month all-in.
MC Alumni Explorer
Type "spinal cord injuries" and get back 13 companies grouped by relevance tier: high, medium, loosely related. Each card shows one sentence of AI reasoning for why it matched. That's the semantic search layer: natural language queries against 4,486 companies, relevance scored and explained. Staff who don't know the sector taxonomy find what they need without knowing what to call it. Staff who do know it get keyword search, eight filter dimensions (sector, sub-sector, funding stage, geography, demographics, confidence score, exit status, program year), card and table views, starred companies, and CSV export alongside.
Each company card shows funding, last funded date, Crunchbase rank, and a survival confidence score from a cohort-adjusted composite model. A button in each detail modal triggers a Claude API call with web search tool use to find and summarize recent press coverage for that company, on demand. One search, one company, live results.
AI-Powered Startup Sector Classifier
Designed and ran an LLM classification pipeline across 4,486 startups: 5 strategic sectors and 5 healthcare sub-sectors. The pipeline feeds elevator pitches and Crunchbase descriptions to Claude Sonnet in batches of 30–40, with structured output parsing, retry logic, and progress tracking. It replaced a keyword-matching approach that had a 26% error rate on reclassifications. That 26% made the database effectively unusable for any segmented analysis.
This is the foundation layer. The Alumni Explorer, the Portfolio Intelligence Dashboard, and four published LinkedIn articles on 15 years of portfolio data all depend on the classified dataset it produced. The order mattered: the classifier had to be right before anything else was worth building.
Portfolio Intelligence Dashboard
A 6-tab interactive dashboard analyzing 15 years of startup outcomes: 4,486 companies, $27.1B in funding, 21 interactive charts. Tabs cover portfolio growth, challenge areas, funding pipeline, survival benchmarks, demographics, and cohort composition. Built as a public-facing data product for VC, corporate innovation, and ecosystem audiences, and as the analytical foundation for four published LinkedIn articles on portfolio data.
The analytical work underneath: a cohort-adjusted survival model calibrated against BLS and Stripe benchmarks (explaining why Crunchbase's raw 87% active rate is actually closer to 59%), a funding pipeline funnel (4,486 companies, 300 Series A+, 12 IPOs), a power law concentration curve (top 1% holds 45% of portfolio funding), and a funding parity analysis by gender and race across all 5 challenge areas.
AI Content Capture Tool with Agentic Research
A web application where marketing and community staff enter information about a startup, mentor, donor, or partner and receive generated marketing copy across four channels: LinkedIn post, blog copy, case study draft, and content angles. Conditional form fields by subject type. Quick-revision buttons (shorter, lead with metric, stronger CTA, warmer tone). Per-output copy-to-clipboard.
The agentic research layer is the differentiator. When a staff member enters a company name, the system autonomously triggers a Claude API call with web search tool use to gather recent press, milestones, and funding data, then simultaneously cross-references an internal alumni database to pull cohort year, sector, and program history. That research context is injected into the generation prompt before copy is produced. Staff get marketing copy informed by live data, not whatever they remember from the intake call.
Tools I actually used.
No scaffolding for the sake of it. Everything here is a deliberate choice for a specific constraint: single-file deploys, no build pipeline, serverless for anything that touches an API key.
The capability I'm building, not the tools themselves.
Being able to ship working tools rather than commission them changes what kinds of bets are feasible and how fast a lean team can move.
The Alumni Explorer took a week. It replaced a workflow that would have taken a quarter to spec and procure. The sector classifier took two days and resolved an error rate that had made the entire database unusable. Crucible went from concept to production in roughly seven Claude Code sessions, and it does work that used to take a program team days per company. Field Notes took a single working session and three rounds of iteration to reach a tool I actually use for every advisory engagement. That's the actual ROI: not the tools, the speed and the scope of what a non-technical operator can now tackle alone.
What's emerged across all seven of these is a pattern. The problems weren't technical; they were operational. Manual reconciliation eating PM time. Gut-feel evaluation without a rigorous framework. A deal intelligence workflow split across three separate products that never talked to each other. AI didn't change what the problems were. It changed how fast one person could address them.