SDE Intern @ Amatyaa · ex AI Developer Intern @ Cubical Frames
Plumbing my way through Backend + AI/ML.🚢
- Voice AI: transcription, diarization, real-time interview system
- Agentic AI: agent tools and workflows
- Production RAG: semantic search, BM25, rerankers, context retrieval, evals
- MCP servers: built and integrated for agent workflows
- Fine-tuning: Whisper large-v3 and BERT-based models
- Backend systems: scalable systems, APIs, cloud and database architecture
-
ragflow (88k★) — #18012: kept short OCR text when no image2text model is configured — the extracted text was being discarded, so those images indexed to zero chunks. Merged.
-
LightRAG (39k★) — #3652: ranked each BFS level by degree before the
max_nodescutoff, so a truncated graph view no longer depended on document ingestion order. Merged. -
marimo (22k★) — #10354: fixed horizontal scrolling when navigating to an off-screen column in columns mode. Merged.
-
pipecat (14k★) — #5141: pushed an
ErrorFrameon in-stream Responses API error events over HTTP — a failed turn used to end in silence, so failover never triggered and nothing was logged. Merged. -
local-deep-research (8.9k★) — #5505: streamed OpenAlex snapshot partitions record by record instead of buffering each one in full — cut peak heap from 607 MB to 12 MB and fixed an OOM on smaller hosts. Merged.
-
WhisperLive (4.2k★) — #531: honored
display_segmentswhen printing transcripts. Merged. -
apache/airflow (46k★) — #70797: restored
/authprefix for FAB Security page links behind subpath proxies. Open. -
supermemory (28.8k★) — #1386: fixed inverted
NEXT_PUBLIC_GOOGLE_AUTH_ENABLED/NEXT_PUBLIC_GITHUB_AUTH_ENABLEDflags so OAuth providers are enabled when set totrue. Open.
Hunting for SWE / AI engineering internships. Real products, people who ship.




