Granth

Creative by night,

more creative

midnight.

20:00

00:00

Coder

AI Engineer

Crafting intelligent Agents

MAIT '29 Delhi, India

WINNER

won MAITRON 2026
in MAIT, Delhi
organised by ISTE

CURIOSITY INTO CREATION.

I've always loved figuring out how things work and imagining what they could become. Today I bring the same curiosity into code, machine learning, and agent AI.

beyond just code.

From training my first ML models to building AI-powered applications, I've learned by making things work.

I've experimented with recommendation systems, RAG chatbots, and multi-agent shopping assistants — breaking things, debugging, and turning rough ideas into projects I'm proud of.

learned a lot
here

SCOUT

less searching, smarter shopping.

SonarIQ

turning what-ifs into what's next

KIRANAMATE

AI that works where businesses already are.

still figuring things out.

I'm an open book.
Here is the unfiltered timeline of how
I figured things out.

how it all started.

SOMEWHERE BETWEEN MY FIRST BOOK AND MY FIRST BREAKTHROUGH, I FOUND MY THING.

As a kid, I spent hours building LEGO houses, dreaming of becoming an architect. Years later, I discovered a different kind of building — coding.

I started with Python and AI in class 11, explored machine learning and deep learning in class 12, and today, I’m turning ideas into intelligent systems.

It started as building
a Lego house to
Building Intelligent Systems

SKILLS

Section 03 // Technical Skills Matrix

Skills

● ONLINE

Interactive Skills Workspace Inspect AI/ML systems, frameworks, languages, and vector databases side-by-side! Drag titlebars to rearrange, view architecture benchmarks, or minimize to taskbar.

+ COORD: 28.6139° N, 77.2090° E // SECTOR_03

+ GRANTH_OS // RETRO_SKILLS_WORKBENCH_v2.6

// APPLE GENIE EFFECT Click any skill to see its architecture emerge with the Genie effect! ↗

🤖 AGENTIC_AI_SYSTEM.EXE // AUTONOMOUS SWARM

FILE AGENT_SWARM REASONING_LOOP TOOL_DISPATCH

AIML_CORE.SYS

AGENTIC AI // MULTI-AGENT SWARMS // PRODUCTION IMPLEMENTATIONS & REPOSITORIES

AI & Machine Learning Core 3 PROJECTS

Multi-Agent Topologies

ReAct Loops

Tool-Use Reliability

// REAL-WORLD PROJECTS WHERE I USED THIS SKILL 3 REPOSITORIES VERIFIED

KiranaMate (MerchantMind AI) Autonomous WhatsApp Swarm

VIEW REPO ↗

// HOW I USED THIS SKILL IN BUSINESS PROJECTS:

Architected 24/7 background autonomous multi-agent loops that ingest Paytm soundbox transactions in real time, detect sales drop anomalies, and autonomously formulate margin-guaranteed WhatsApp buyer campaigns without needing manual merchant prompts.

// WHAT I'M DOING NEXT WITH THIS SKILL:

Integrating voice-first autonomous agent negotiation in 10+ Indic vernacular languages with real-time audio streams.

Multi-Agent StateGraph WhatsApp Cloud API Sarvam AI FastAPI

healthcare-guardrails-agent Layered Guardrails & HITL

VIEW REPO ↗

// HOW I USED THIS SKILL IN BUSINESS PROJECTS:

Engineered deterministic pre-LLM safety filters, PII redaction middleware, and Human-in-the-Loop approval checkpoints before invoking sensitive appointment-booking and medical tools.

// WHAT I'M DOING NEXT WITH THIS SKILL:

Implementing automated red-teaming adversarial agent probes to dynamically stress-test safety boundaries before clinical deployment.

LangChain LangGraph PII Middleware Streamlit

SCOUT Omnichannel Shopping Swarm

VIEW REPO ↗

// HOW I USED THIS SKILL IN BUSINESS PROJECTS:

Built autonomous decentralized agent swarm coordinating parallel product discovery, inventory price tracking, and multi-platform consumer search across Indian e-commerce platforms.

// WHAT I'M DOING NEXT WITH THIS SKILL:

Adding multi-agent auction bidding heuristics and distributed web crawler agent nodes for live flash sales.

Autonomous Agents Tool Dispatch Dynamic Routing

// SYSTEM SPECIFICATION & ARCHITECTURE BENCHMARKS STATUS: VALIDATED

Module / Pipeline	Technology	Benchmark / Specs	Operational Role
Agentic Swarm Topologies	LangGraph + Multi-Agent	Cyclic Graphs	Parallel routing, inter-agent consensus & task delegation
Reasoning & Reflection	ReAct + Chain-of-Thought	Zero Drift	Self-evaluating validation nodes and error recovery loops
Tool Dispatch Engine	Structured Function Calling	Pydantic Schemas	Strict type-checked IO, rate limits & external API bridges
Deep & Reinforcement Learning	Policy Gradients & Networks	Convergence	Reward formulation, exploration strategies & agent policies

C:\SKILLS\AI_ML\AGENTIC_SYSTEMS READY // 0 ERRORS

🕸️ LANGGRAPH_STUDIO.IO // STATEGRAPH ENGINE

FILE STATE_GRAPH CHECKPOINTERS LANGSMITH_TRACES

TRACING: 100% CAPTURED // GRAPH SYNCED

LANGGRAPH & LANGSMITH // STATEFUL CYCLIC GRAPHS // PRODUCTION IMPLEMENTATIONS & REPOSITORIES

Frameworks & Orchestration 3 PROJECTS

Cyclic StateGraphs

100% Trace Coverage

HITL Checkpoints

// REAL-WORLD PROJECTS WHERE I USED THIS SKILL 3 REPOSITORIES VERIFIED

healthcare-guardrails-agent StateGraph & Resumable Checkpoints

VIEW REPO ↗

// HOW I USED THIS SKILL IN BUSINESS PROJECTS:

Developed stateful LangGraph StateGraph with MemorySaver checkpointers for resumable human approval workflows, conditional routing edges for medical output validation, and LangSmith distributed tracing for token spend and latency auditing.

// WHAT I'M DOING NEXT WITH THIS SKILL:

Deploying persistent PostgreSQL checkpointers with time-travel workflow rollbacks and state diff visualizers.

LangGraph StateGraph MemorySaver LangSmith Traces Pydantic

Twtr_post_generator (tweet_forge) Cyclic Self-Reflection Graph

VIEW REPO ↗

// HOW I USED THIS SKILL IN BUSINESS PROJECTS:

Designed cyclic multi-persona agent loop (Generator -> Evaluator -> Optimizer) with conditional branching edges; iterates and self-corrects drafts until meeting editorial approval or max iterations.

// WHAT I'M DOING NEXT WITH THIS SKILL:

Adding dynamic multi-modal preview generation and real-time Twitter API performance metric feedback loops.

LangGraph Cyclic Loops TypedDict State Conditional Routing Groq

KiranaMate Multi-Agent Workflow Orchestration

VIEW REPO ↗

// HOW I USED THIS SKILL IN BUSINESS PROJECTS:

Orchestrated complex proactive decision graphs coordinating anomaly detection, competitor price verification, and multi-channel notification execution with state persistence.

// WHAT I'M DOING NEXT WITH THIS SKILL:

Parallel multi-branch execution with asynchronous sub-graph spawning.

LangGraph Workflows Dynamic State Error Fallback Nodes

// SYSTEM SPECIFICATION & ARCHITECTURE BENCHMARKS STATUS: OBSERVED

Module / Pipeline	Technology	Benchmark / Specs	Operational Role
Graph Workflow Compiler	LangGraph StateGraph	Cyclic Topology	Compiled state schemas with conditional edges and branch routers
Persistence & Checkpointers	MemorySaver / Postgres Checkpointer	Instant Rollback	Session resumption, human-in-the-loop pauses, and state diffing
Observability & Monitoring	LangSmith SDK	Live Traces	Distributed span tracking, token spend auditing & latency bottlenecks
Evals & Test Suites	LangSmith Datasets	Regression Guard	Automated prompt evaluation against gold standard benchmark suites

C:\SKILLS\FRAMEWORKS\LANGGRAPH_ORCHESTRATION VIEW REPO ↗

🔍 RAG_MULTIMODAL_PIPELINE.SYS // HYBRID RETRIEVAL

FILE RETRIEVER VECTOR_SPACE RERANKER

PIPELINE: GROUNDED // <50MS RETRIEVAL

RAG PIPELINES & MULTIMODAL RAG // PRODUCTION IMPLEMENTATIONS & REPOSITORIES

Retrieval-Augmented Generation 3 PROJECTS

Hybrid Dense + Sparse

<50ms FAISS Latency

Zero Hallucination

// REAL-WORLD PROJECTS WHERE I USED THIS SKILL 3 REPOSITORIES VERIFIED

RAG_yt_chatbot Sub-50ms Transcript Vector RAG

VIEW REPO ↗

// HOW I USED THIS SKILL IN BUSINESS PROJECTS:

Built end-to-end RAG pipeline extracting YouTube transcripts, chunking with sliding context windows, indexing with Sentence Transformers into a FAISS vector database, and generating cited context-aware answers using Groq Llama 3.3 70B in under 50ms.

// WHAT I'M DOING NEXT WITH THIS SKILL:

Implementing multi-video comparative synthesis with hierarchical graph-guided document indexing.

FAISS Sentence Transformers LangChain Groq Llama 3.3 70B

SatQuery-AI Multimodal Remote Sensing RAG

VIEW REPO ↗

// HOW I USED THIS SKILL IN BUSINESS PROJECTS:

Engineered multimodal vision-language RAG pipeline pairing satellite remote sensing embeddings with natural language query retrieval for flood assessment, temporal change detection, and spatial analytics.

// WHAT I'M DOING NEXT WITH THIS SKILL:

Expanding to multi-spectral temporal change diffing with real-time geospatial bounding queries.

Vision-Language Models Multimodal Embeddings Geospatial Analytics

KiranaMate Knowledge Graph Memory Retrieval

VIEW REPO ↗

// HOW I USED THIS SKILL IN BUSINESS PROJECTS:

Implemented Cognee knowledge graph memory for semantic merchant retrieval, anchoring autonomous agent decisions in historical store transactions to eliminate hallucinations.

// WHAT I'M DOING NEXT WITH THIS SKILL:

Continuous episodic memory consolidation across multi-month sales histories.

Cognee Memory Knowledge Graphs Semantic Indexing

// SYSTEM SPECIFICATION & ARCHITECTURE BENCHMARKS STATUS: OPTIMIZED

Module / Pipeline	Technology	Benchmark / Specs	Operational Role
Hybrid Retrieval Engine	Dense Vectors + BM25	RRF Reranking	Blends semantic meaning with precise keyword and code recall
Multimodal Ingestion	Vision LLMs + OCR Parsing	Unified Vectors	Extracts diagrams, tabular data, and schematics into embeddings
Vector Index Optimization	FAISS (HNSW / IVF-PQ)	Sub-millisecond	High-dimensional similarity indexing with memory quantization
Context Synthesis	Reciprocal Rerank + Citations	Strict Grounding	Eliminates token bloat and synthesizes verbatim cited responses

C:\SKILLS\RAG\MULTIMODAL_RETRIEVAL VIEW REPO ↗

19:55:00

Brand Design Social Media Typography Poster Design Colour Grading Motion Graphics Visual Identity Content Creation

insomniac Work

contact Me

Ready to make a move? Drop an email to discuss
internships, collaborations, or just to say hi.

→ say hi before overthinking it