S Sai Kumar

LLM & Agentic AI

// Engineering student building intelligent systems with LLMs, agents, and full-stack craft

ask the agent anything

/PROJECTS

argus-v — flagship.exe

Name

A.R.G.U.S-VVideo intelligence framework for YouTube

Synopsis

argus-v --stream <source> --query "<question>"

Description

Multimodal video-intelligence system with RAG and LangGraph agents — streams YouTube video through parallel ASR, scene detection, and vision-embedding pipelines into a queryable knowledge base.

Results

  • Streaming ingestion unlocks natural-language querying ~13–20s after a video starts processing, instead of minutes
  • LangGraph agent with dual-LLM routing; replacing an LLM critique loop with rule-based evaluation cut ~28s per query
  • 8-GPU pipeline on DGX H200: FastAPI ASR + vision microservices, Redis task queue, LanceDB vector store
  • Python
  • LangGraph
  • vLLM
  • Whisper/Canary ASR
  • Qwen3-VL
  • LanceDB
  • Redis
  • FastAPI
Read full case study →
monetary-policy — research.ipynb

In [1]markdown

Voice of Monetary Policy

Multimodal analysis of FOMC meetings

In [2]code

fusion = concat(
  bert_sentiment(fomc_transcripts),   # hawkish / dovish signal
  prosody_features(fomc_audio),       # pitch, tempo, stress
)
bootstrap_ols(fusion, market_reaction, n_boot=2000, ci="BCa")

Out [2]result

F1 0.82

Replication of Gorodnichenko, Pham & Talavera (2023): does the Fed's tone of voice move markets independently of its words? Fuses vocal prosody (speech-emotion recognition) with BERT text sentiment across 792 FOMC segments, evaluated with bootstrap OLS on 22 financial instruments.

Read full case study →
openwell — daemon.service

openwell.service — Digital mental-health platform

Loaded: loaded (/etc/systemd/system/openwell.service; enabled)Active:active (running)Docs: openwell.vercel.app

Full-stack mental-health support platform: PHQ-9/GAD-7 self-assessments, an AI support chatbot with contextual risk assessment, counsellor booking, and anonymous access — localized in English, Hindi, Kashmiri, and Dogri.

/SYSTEM

skills — htop
ProcessLoad

Languages · Python / C/C++ / Java / JavaScript

AI & ML · LLMs / Agentic AI / LangGraph / Machine Learning / Deep Learning / Transformers / OpenCV

Web & Tools · React / Node.js / MERN stack / Redis / Git / Claude Code / HTML/CSS

contributions — cal.sh

$ ./cal.sh --last 12mo

> 236 commits in the last 12mo · private work not shown

lessmore

/WHOAMI

whoami — education.log

[2020]INFOSecondary Education → Kendriya Vidyalaya Hirapatti, Azamgarh, Uttar Pradesh

[2022]INFOHigher Secondary Education → Kendriya Vidyalaya No. 1, Akhnoor, Jammu & Kashmir

[2027]WARNB.Tech in Computer Science (AI & Data Science) → SASTRA Deemed University, Thanjavur, Tamil Nadu — expected 2027, pending final semester

[CERT]INFODSA to Web Development — GeeksForGeeks

[CERT]INFOCloud Computing — Coursera

contact — mailto.sh
$ echo $EMAIL →s.sai08019@gmail.com
$ open github →github.com/SaiK1105
$ open linkedin →www.linkedin.com/in/saik7337
$ open tel →+91 7337354430