Folders and files
| Name | Name | Last commit date | ||
|---|---|---|---|---|
AI Engineer · Production LLM Systems · Agentic Workflows · Future Standard · Bangalore
I build production AI systems end-to-end — from database to infra to deployed LLM pipelines. My focus is agentic systems that hold up in production: memory architectures, RAG over hard domains, evals, and orchestration.
Currently Analyst — AI & Agent Systems at Future Standard, designing agentic memory architectures and building agent systems for operations and process intelligence — orchestration (LangGraph/LiteLLM), backend APIs, evaluation pipelines, and production deployment.
Previously at OnFinance AI, where I built ComplianceOS end-to-end (DB → backend → cloud → LLM pipelines) for BFSI compliance automation — production RAG over financial regulations with versioned embeddings and citation grounding, LangGraph + LiteLLM agentic workflows processing 1B+ tokens/month, cutting compliance effort by 1,500+ hours/year. Ex-RingCentral (SDE intern — Angular → React modernization, 3x faster load times, 95% test coverage).
- BE in Computer Science @ RV College of Engineering, Bangalore · former President, Coding Club RVCE
- Contributor to maximhq/bifrost (3.4k★) — fixed streaming correctness bugs in the AWS Bedrock provider
- Working across the LLM stack: agentic workflows (LangGraph), RAG pipelines (ChromaDB, Ollama), multi-model evaluation, TTS, image/video generation
- Interested in agent memory, LLM infrastructure, distributed systems, and developer tooling
maximhq/bifrost — Enterprise AI Gateway · 3.4k★ · Go · ~11µs overhead at 5,000 RPS · 20+ LLM providers
| PR | Description |
|---|---|
| #2427 | Fixed Bedrock streaming silently ignoring max_retries — transport errors (*net.OpError, io.ErrUnexpectedEOF, *net.DNSError) routed through wrong error path, bypassing retry gate; added isStreamTransportError() helper across 4 sites + 3 regression tests |
| #1830 | Fixed concurrent tool call delta loss — deltas routed by recency instead of tool_call.id, causing parallel call arguments to merge; fix applied to both Chat Completions and Responses API streaming paths |
| Project | Description | Stack |
|---|---|---|
| faceless-gen | Local-only AI video studio — research doc in, MP4 out. 5-step editorial wizard (angles → story → scenes → music → generate). Stickman style via ComfyUI + Wan 2.2 I2V. 218 tests passing. | |
| graph-rca | Log-to-resolution RAG over 200 annotated real-world incidents — LLM builds causal DAGs, ChromaDB (HNSW) retrieves runbook context, output quality validated by a 3-model judge ensemble (Qwen3:32b, GPT-4o-mini, Llama-3.1-70B) across 9 reproducible experiment scripts; GPU-accelerated, includes React frontend | |
| permission-mongo | RBAC-powered BaaS for MongoDB targeting 50K+ QPS — AST-caching policy compiler, lock-free router, fasthttp, Prometheus + Grafana dashboard, Python SDK | |
| better_bing_image_downloader | Maintained Python library · 18K+ PyPI downloads · 41★ · v3.0.0 fixed thread-safety, atomic writes, MD5 dedup; test suite grew 0 → 41 tests | |
| snapshell | CLI tool to learn Linux commands interactively | |
| toy-compiler | Full-pipeline compiler (lexer → parser → codegen) for a subset C-like language |
Languages
Backend & Messaging
LLM & AI Tooling
Infra & Observability
Databases