technical architecture
Reducing Data Pipeline Latency to Enable Real-Time Decision Loops
Redesigned batch pipeline to streaming architecture, reducing data lag from 2-3 days to <2 hours and enabling new decision workflows
Work Artifacts
Case Studies
Overview
These case studies demonstrate product judgment, technical depth, and execution leadership from my tenure at Montai, a drug discovery AI startup. Each follows the format: Context → Ownership → Decision Frame → Outcome → Reflection.
Focus: Decision systems, not achievements. What options existed, what I chose, why, and what I learned.
Featured Case Studies
Scaling AI-Driven Drug Nominations
Product Strategy + Technical Architecture | 2023-2024
Scaled compound nominations 26× (250 → 6,500+ per program) while improving hit-to-lead rates from 5% to 27%. Built end-to-end data pipeline automating ML predictions, designed phased rollout strategy balancing speed and quality.
Key decisions: Phased scaling (prove → scale → refine) vs immediate optimization, quality gates to prevent stakeholder trust erosion, balancing exploration and exploitation.
From Data Crisis to Data Culture
Execution Leadership + Incident Response | 2025
Led response to critical data integrity failure (STAT6 predictions missing, 6-week program delay). Established postmortem process and data governance framework preventing recurrence (3 incidents in 2024 → 0 in 2026).
Key decisions: Targeted fix + governance uplift vs quick patch or comprehensive rebuild, blameless postmortem culture, proactive monitoring investment post-crisis.
Learning Agendas: Research Rigor for Product Decisions
Product Strategy + PhD Transfer | 2025
Designed decision framework reducing R&D cycle time 20% (~10 weeks → ~8 weeks) by pre-defining success criteria and pivot triggers. Applied academic experimental design to product/strategy decisions.
Key decisions: Lightweight one-page format vs heavyweight docs, enforcing pre-commitment to decision criteria, balancing rigor and pragmatism for scientist adoption.
Build vs Buy: Strategic Analysis for Analog Generation
Strategic Analysis + Executive Communication | Q4 2025
Quantitative analysis guiding $250k+ partnership decision: XtalPi external compounds vs improving internal generative model. ROI modeling revealed internal model needed 50× accuracy improvement; recommended hybrid approach (external for near-term + internal investment for long-term).
Key decisions: Build vs buy rarely binary (hybrid optimal), quantitative framing transforms opinions into evidence, strategic patience requires near-term wins.
Standardizing Montai’s App Ecosystem with R Shiny
Technical Architecture + Developer Experience | 2024
Converged fragmented app development (Python/Streamlit/Dash + R/Shiny + notebooks) onto single framework, cutting development time 66% (3 weeks → 1 week). Built proof-of-concept Nomination App and reusable component library.
Key decisions: R Shiny (team skills + data-heavy use case) vs Python frameworks vs multi-framework flexibility, trading long-term flexibility for near-term velocity, standardization as social + technical choice.
Additional Case Studies
Preventing Metric Theater in Drug Discovery ML
Product Strategy + Evaluation Frameworks | 2024
Designed combined offline + online evaluation framework de-risking $2M+ compound decisions by catching model degradation 3 months earlier. Established monitoring infrastructure reused by 3+ model teams.
Reducing Pipeline Latency for Decision Velocity
Technical Architecture + Data Engineering | 2023-2024
Automated data pipeline reducing latency from 2-3 days to same-day updates. Enabled real-time analytics dashboards used for investor updates and program decisions.
Themes Across Case Studies
Decision Systems Over One-Off Analyses
- Learning Agenda framework (reusable structure, not one project)
- Build vs Buy quantitative analysis (template for future decisions)
- Data Quality governance (postmortem process, not one incident fix)
Product Judgment + Technical Depth
- Strategic analysis (XtalPi ROI modeling) + technical execution (Shiny framework)
- Quality gates (nomination filtering) + pipeline automation (dbt implementation)
- PhD rigor (experimental design) + startup pragmatism (lightweight formats)
Candid Reflection on Tradeoffs
- Phased scaling delays (perfect vs done tension)
- Framework lock-in risks (Shiny standardization)
- Crisis response delays (monitoring investment should’ve been proactive)
product strategy
Preventing Metric Theater in Drug Discovery ML
Designed evaluation framework combining offline validation with online performance tracking to de-risk multimillion-dollar compound selection decisions
product strategy
Build vs Buy: Strategic Analysis for Analog Generation
Quantitative analysis guiding $250k+ partnership decision — ROI modeling revealed internal model needed 50× improvement, recommended hybrid approach balancing near-term progress with long-term IP
product strategy
Learning Agendas: Bringing Research Rigor to Product Decisions
Designed decision framework cutting R&D cycle time 20% by pre-defining success criteria and pivot triggers — applied academic experimental design to product strategy
execution leadership
From Data Crisis to Data Culture: The STAT6 Incident
Led 6-week crisis response resolving data integrity failure — transformed incident into governance framework reducing future incidents from 3 to 0
technical architecture
Standardizing Montai's App Ecosystem with R Shiny
Converged fragmented app development onto single framework, cutting development time 66% — balanced team skills with deployment simplicity to accelerate internal tool velocity
technical architecture
Scaling AI-Driven Drug Nominations from 250 to 7,000 Compounds
Built pipeline scaling compound nominations 26× while improving hit rates from 5% to 27% — phased rollout strategy balanced speed with quality gates