
I fix the bugs that never throw an error.
Contact
Harbour Square Park, Toronto
I’m a software developer in Toronto. I work on backends, cloud systems and AI agents, and I contribute fixes to the open-source tools engineers rely on.
The bugs I like most never raise an error. They return slightly wrong data, or send a pipeline down the wrong branch after midnight, and nobody notices for months.
Some leave no trace at all. The one here is real: every time someone closed a log view in Argo CD, the server quietly kept a little more memory it would never give back. More about me
Try the Argo CD log leak
Closing a pod log view while logs were still streaming left one goroutine stuck in argocd-server, forever. My fix is merged and was cherry-picked into the 3.5 and 3.6 releases.
- argocd-server, before the fix3 goroutines stuck
Each one is blocked forever on its second send to
done, still holding its log stream.- argocd-server, after the fix0 goroutines stuck
The goroutine reports the error and returns.
}); err != nil { done <- err- break+ return}Fixes in open source
14 fixes across 9 projects, each written up with the problem, the fix and how I tested it.
Making XCom slicing match Python semantics across every databaseApache Airflow
Fixing overnight time windows in BranchDateTimeOperatorApache Airflow
Correct weekly bucketing on Hive, with real-database CI coverageApache Superset
Keeping 0 and false filter values in Jinja templatesApache Superset
Fixed Monday week start for Oracle and ImpalaApache Superset
Eliminating a goroutine leak in pod log streamingArgo CD
Correct partition pruning for smallint columnspg_lake
Keeping filtered parameters out of request logsRuby on Rails
Returning the requested offset from TimeWithZone#changeRuby on Rails
Making streamed responses safe for shared header mapsGin
Preventing arguments from silently changing kubectl’s log levelKubernetes
Consistent label renaming in the logfmt parserGrafana Loki
Restoring AsyncAPI generationFirestone
Applying resource-wide security to every endpointFirestone

Civic Tech Toronto
Volunteer developer since September 2026, on Civic Dashboard.
Civic Tech Toronto began on 6 June 2015, when a group of residents spent a day working on problems in their own neighbourhoods. Since that summer it has met every Tuesday evening, in person and online, and it has now passed its 500th hacknight. It is run entirely by volunteers who want to understand and fix civic problems with technology.
I volunteer on Civic Dashboard, one of its projects, which makes Toronto City Council easier to follow. Residents can see what is coming up, read their councillor’s voting record, get an email when an item they care about is on the agenda, and send a comment or ask to speak in one step.

What I’ve fixed
Searches in brackets matched everything
A search for (housing) returned every item, and email alerts use the same parser, so subscribers got an email for every new item. Brackets now group words and a minus hides the whole group.
Merged · #549
The feedback form could ping a whole Slack channel
Form fields went into the team’s Slack messages as written, so anyone could trigger @channel or post a disguised link. Every field is now escaped and shows as plain text.
Merged · #557
Pages that repeated or skipped items
Infinite scroll could show an item twice or miss one, a failed unsubscribe still hid the subscription, and the public APIs accepted any page size. Four small fixes, one per bug.
Merged · #559
Agenda pages that would crash in the new council term
An upcoming item whose committee was unknown returned a server error, which would hit every item once the new term began. The page now loads and hides only the actions it cannot send.
In review · #570
A security upgrade that would have taken the site down
Upgrades Next.js to close a high-severity denial-of-service advisory, together with the Cloudflare adapter versions that keep every page working on the new release.
In review · #572
Hackathons
FalconX
2nd prize at the TechTO hackathon, 17 October 2026
An AI voice agent that does a switchboard’s job: it phones several hotels at once, checks the real rate and books the cheapest.
Read the case studyContinuum
Built for Devpost’s OpenAI Build Week, 2026
A ledger of everything an AI session did, so the next session can pick up exactly where the last one stopped.
Read the case study





Projects
Nine things I built, from e-commerce backends to the tools that make shipping safer.
- AI Job Search AssistantFull-stack / AI2026

- RSC Storybook BridgeFrontend Platform / DX2026

- React 19 Migration Safety NetPlatform / DX2026

- SSE Streaming Reliability KitBackend / Perf2026

- ContinuumPlatform / AI Tooling2026

- VibeCartFull-stack2025

- ShopSphere Microservices E-CommerceBackend / Microservices2025

- Fragments APIBackend2024

- FalconXProduction / AI Voice2026


What I build with
- Languages
- TypeScript
- Python
- Go
- Java
- Ruby
- SQL
- AI and agents
- LLM agents
- Model Context Protocol
- Voice agents
- Gemini on Vertex AI
- Tool calling
- Structured outputs
- Backend and data
- Node.js
- FastAPI
- Spring Boot
- Fastify
- PostgreSQL
- Redis
- Kafka
- DynamoDB
- Cloud and testing
- AWS
- Google Cloud
- Docker
- Kubernetes
- Terraform
- GitHub Actions
- Vitest
- Playwright

About me
I build software that holds up in the real world: AI agents that take real actions, backends that stay fast under load, and the tooling that keeps engineering teams moving.
Most of my work sits where AI meets production systems. FalconX, an AI voice agent that phones hotels to verify and book rates, took 2nd prize at TechTO. Continuum gives AI tools a shared, verifiable memory across sessions and models.
I care about the details most people never see until they break: a race condition, a time zone edge case, a filter that silently drops a value. That is usually where the interesting work is.
I studied Computer Programming and Analysis at Seneca Polytechnic, and I live and work in Toronto.
Dhruv Patel
Software developer, Toronto

I’m always up for a hard problem, an ambitious idea or a good collaborator.



