Skip to content
View sanjibani's full-sized avatar

Block or report sanjibani

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.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. 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
sanjibani/README.md

Β  Β  Β 


πŸ‘‹ Who Am I?

I'm a senior software engineer with 10+ years of experience, originally trained in Electronics & Communication Engineering β€” which means I've always been fascinated by the full stack: from signals to systems to the software built on top of them.

Today I focus on Agentic AI systems β€” building multi-agent pipelines that do real-world research, forensic analysis, and automation, not just chat. I particularly love the engineering challenge of making LLM-powered systems reliable, not just impressive.

When I'm not shipping code, I teach high school mathematics and enjoy breaking down complex technical concepts into things anyone can understand. That same instinct drives my writing on Substack and my YouTube channel.

🎯 My superpower: I can navigate between a low-level circuit diagram and a high-level LLM orchestration graph without losing context. Rare combo.


πŸš€ Featured Projects

πŸ•΅οΈ ForensicValue AI

Multi-agent forensic accounting for Indian stocks

Detects accounting irregularities, governance risks & value traps in NSE/BSE companies. Five parallel agents (Forensic, Management, RPT, Market Intelligence, Narrative) orchestrated via LangGraph, synthesized into a "Detective Story" report.

Python LangGraph Streamlit Gemini/Claude

Daily autonomous research pipeline for special-situation investing

Monitors NSE/BSE filings for insider trades & corporate events, then runs a 4-level deep-research pipeline (DeepSeek V3 + Qwen). Persists insights in PostgreSQL with pgvector. Feedback loop trains a reward model to personalize over time.

Python LangGraph PostgreSQL pgvector Redis Docker

AI that reads your logs so you don't have to β€” at 3 AM

Ingests production logs, clusters them with HDBSCAN, flags anomalies with Isolation Forest, investigates root causes via cross-encoder reranked RAG, and drafts fixes for human approval. Human-in-the-loop learning improves recommendations over time.

Python HDBSCAN ChromaDB OpenAI Kafka Streamlit

AI-powered drug information assistant

A conversational tool for looking up medicines, active ingredients, and substitutes β€” built with a clean, fast React frontend and an intelligent backend that surfaces the right information quickly and safely.

React Node.js LLM


πŸ› οΈ Tech Arsenal

Languages & Runtimes

Python JavaScript TypeScript C

Frontend

React Next.js

AI / ML

LangChain LangGraph HuggingFace OpenAI

Infra & Data

Docker PostgreSQL Redis Git


πŸ“Š GitHub Stats


✍️ What I'm Creating

⚑ This section auto-updates via GitHub Actions β€” check back for my latest Substack posts and YouTube videos.


🧠 Currently Exploring

  • Building more reliable agentic systems (the harness matters as much as the model)
  • Forensic accounting automation for Indian public markets
  • Low-level systems β€” revisiting the ECE roots
  • Value investing through the lens of AI-assisted research

Popular repositories Loading

  1. VocabQuest VocabQuest Public

    TypeScript 1

  2. value-investing-agents value-investing-agents Public

    Python 1

  3. Detectron Detectron Public

    Forked from facebookresearch/Detectron

    FAIR's research platform for object detection research, implementing popular algorithms like Mask R-CNN and RetinaNet.

    Python

  4. ReactCode ReactCode Public

  5. ReactiveForm ReactiveForm Public

  6. angular-v2h3xm angular-v2h3xm Public

    Created with StackBlitz ⚑️

    TypeScript