What changed, what made it difficult, and what shipped. Founder work is labeled separately from professional experience.
Founding engineer and cofounderV2 beta live
Rebuilding a paid networking product around structured professional proof.
GreenRoom V1 validated paid artist consultations. V2 needed a safer data model, stronger discovery, and a migration path that preserved the trust already earned.
130+legacy accounts migrated
137paid meetings preserved
49Postgres tables across four graph layers
Natural-language goals resolve against structured profiles, credits, relationships, momentum, and intent.
The constraint
V1 stored identity, payments, and booking history across AWS serverless services. A clean rewrite could not erase users, completed consultations, or authentication continuity.
The engineering move
I designed a four-layer Postgres graph model for entities, relationships, momentum, and intent, then built a DynamoDB to Postgres ETL and Firebase authentication bridge.
Why it matters
The product can now answer higher-value discovery questions while keeping legacy customers and transactions intact. AI operates over structured evidence rather than ungrounded profile text.
Schematic of the executed V1 → V2 migration. Rollback stayed cheap until seven consecutive drift-free reconciliation nights promoted V2.
Published work becomes structured profile evidence during onboarding.Credits remain typed, searchable, and tied to supporting sources.
See the V2 product logic in three real screens, or visit the public V1 while the beta runs with current users.
Reducing data onboarding and model calibration from operational bottlenecks to repeatable systems.
85%faster new data-source onboarding
100M+parameter search space distributed
2 hrscalibration runtime after redesign
The system
A self-service Snowflake framework automated schema creation, access controls, and ingestion templates for new sources.
The scale problem
A Python, Kafka, and AWS ECS calibration system distributed grid search that previously ran locally for days or weeks.
The result
Analysts gained repeatable access to firm-wide credit datasets, while model calibration completed in roughly two hours.
Schematic reconstruction from public resume material. Internal interfaces and proprietary data are not shown.
Professional experience summarized from public resume material. Internal interfaces and proprietary data are not shown.
Bloomberg
Software engineer, Data Technologies
Moving high-volume financial data without slowing the research workflows built on top of it.
6,500+companies in the ESG migration
45%lower query latency
Globalfinancial data acquisition and delivery
The work
I engineered a Kafka migration for ESG company data used in quantitative research and built normalized bank-filing pipelines for Terminal credit analysis.
The constraint
Data contracts and downstream research behavior had to remain dependable while acquisition and processing systems changed underneath them.
The result
The migration reduced query latency by 45 percent while continuing to serve workflows built around thousands of companies.
Schematic reconstruction from public resume material. Bloomberg client systems and proprietary data are not shown.
Professional experience summarized from public resume material. Bloomberg client systems and proprietary data are not shown.
QuantMechanix
Independent research platformDecision support
Making a trading research pipeline resumable, explainable, and hard to bypass.
This is a strategy-research and decision-support platform, not a claim of investment performance.
500 to 30 to 10daily symbol-reduction pipeline
13model ensemble in the current source of truth
9blocking pre-trade checks
The desk compresses a broad daily scan into a smaller review queue with model context and explicit risk checks.
The constraint
Market-data limits, partial failures, and long-running analysis made an all-or-nothing batch unreliable and difficult to audit.
The engineering move
I added Postgres checkpoints, Redis rate limiting, scheduled ingestion, prediction logging, and automated end-of-day outcome settlement.
The risk control
A server-stamped PASS or FAIL gate blocks execution unless all nine checks succeed. The interface exposes uncertainty instead of hiding it.
Every stage is resumable from its checkpoint; nothing reaches execution without all nine server-side checks passing.
Conviction remains visibly unproven until the evidence threshold is met.The queue separates trade-ready research from watch and reject states.
Continue the conversation
Building or hiring in this territory?
For GreenRoom, a senior engineering role, or focused systems work, email is the fastest path.
