Skip to content
View astroknot-sheep's full-sized avatar

Block or report astroknot-sheep

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
astroknot-sheep/README.md

Astroknot sheep, a hand-drawn orbital illustration and live upstream pull request numbers

hi, i'm dhriman.

i studied economics at IISER Bhopal and ended up working on machine learning systems. lately i've spent a lot of time in other people's repositories. i like finding out why something broke, fixing it, and seeing whether the fix survives review.

and yes, the sheep in @astroknot-sheep was deliberate. 🐑

pull requests · my repositories · discussions


some fixes that made it upstream

these are the kind of contributions i enjoy: specific bugs, readable changes, and a review thread where someone makes the code better.

  • MLflow — Gemini safety blocks weren't being reported as content-filter events. fixed how the gateway handles them.
  • Qdrant Client — the default tokenizer for a gRPC text index wasn't matching the expected word default.
  • Haystack — corrected line ranges when PythonCodeSplitter makes secondary splits.
  • Logfire — closing a generator early shouldn't look like an application error.
  • Sentence Transformers — fixed a broken training-arguments import in the examples.

there are more. the actual code and discussions are linked, so you don't have to take my word for any of it.

pull requests, over time

A chart showing public upstream PRs opened and merged each month

where i've contributed — by repository Repositories sorted by the number of PRs I've opened

Days I opened PRs during the past year

the numbers above count public PRs to repositories i don't own. PRs to my own repos aren't mixed in. data comes from GitHub and refreshes daily; this heatmap is about PRs, not commits.

merged PRs — with dates and times

this is the complete upstream merge history, newest first. the timestamps come from GitHub's actual merge events, not the date the PR was closed.

Merged at (UTC) PR Repository Time to merge
16 Nov 2026, 12:35 UTC fix: convert common Python types in Agent tool results to JSON deepset-ai/haystack 3.5 days
16 Nov 2026, 10:03 UTC Report the reducer in unified TS.MRANGE/MREVRANGE metadata on RESP2 with GROUPBY redis/redis-py 3.6 days
13 Nov 2026, 07:44 UTC fix(pgvector): cast numeric meta filters to numeric instead of integer/real deepset-ai/haystack-core-integrations 22.4 days
13 Nov 2026, 05:37 UTC fix(docx): close an open list before a table that follows it docling-project/docling 1.2 days
12 Nov 2026, 10:19 UTC Don't record GeneratorExit as an error when a generator is closed early pydantic/logfire 18.8 hours
12 Nov 2026, 07:55 UTC fix: report per-piece line ranges for PythonCodeSplitter secondary splits deepset-ai/haystack 17.2 days
26 Oct 2026, 09:33 UTC [examples] Fix SentenceTransformerTrainingArguments import in paraphrase and multi-task examples huggingface/sentence-transformers 4.6 days
24 Oct 2026, 04:15 UTC Map Gemini safety blocks to content_filter in the gateway mlflow/mlflow 1.6 days
22 Oct 2026, 06:44 UTC fix: default the text index tokenizer to word in grpc create_payload_index qdrant/qdrant-client 8.4 hours

recent PRs

Opened Pull request Repo Status
2026-11-16 fix(sqlalchemy): keep colons in SQL string literals literal deepset-ai/haystack-core-integrations open
2026-11-13 fix: convert common Python types in Agent tool results to JSON deepset-ai/haystack merged
2026-11-12 Support @logfire.instrument above @classmethod and @staticmethod pydantic/logfire closed
2026-11-12 Report the reducer in unified TS.MRANGE/MREVRANGE metadata on RESP2 with GROUPBY redis/redis-py merged
2026-11-12 fix: pass the content type to the deserializer in LocalEndpoint.invoke aws/sagemaker-python-sdk open
2026-11-12 fix(docx): close an open list before a table that follows it docling-project/docling merged
2026-11-11 Don't record GeneratorExit as an error when a generator is closed early pydantic/logfire merged

all pull requests · merged upstream · still open


things i'm making

KONARQ_MATRI
a mental-health support project. i'm thinking about privacy, referrals, and where software should step aside for a human.

finance_transformer
trying transformer models on financial time series. mostly a lesson in how convincing noise can look.

DQN for Ludo
double DQN, prioritized experience replay, and a game that makes the experiments less abstract.

older experiments and side projects

the usual tools

Python, PyTorch, TypeScript, Git, Docker, Jupyter. depends what i'm breaking that week.

say hello

have a question about a PR? spotted a mistake? working on something similar?

you can open a discussion. i don't mind questions, including beginner ones.

this profile's graphs are generated in-repo with Python and GitHub Actions. source code.

Popular repositories Loading

  1. finance_transformer finance_transformer Public

    We try to implement famous deep learning architecture "transformer" with changes to handle continuos non-discrete financial data to observe,experiment and learn.

    Python 1

  2. Reinforcement_Learning_DQN_for_Ludo Reinforcement_Learning_DQN_for_Ludo Public

    Trying DQN for household Indian game Ludo.

    Python 1

  3. Parfum Parfum Public

    Python 1 1

  4. KONARQ_MATRI KONARQ_MATRI Public

    TypeScript

  5. ML_project_final ML_project_final Public

    Jupyter Notebook

  6. mlproject mlproject Public

    Jupyter Notebook