AI Engineer · System Architect · Research Enthusiast
Manan Panchal
Building intelligent systems at the intersection of
|
Precision over
noise.
I'm Manan Panchal — a third-year B.Tech AI & ML student at CHARUSAT, Gujarat, and an SSIP Grant recipient. I build production-grade ML systems that bridge research and real-world impact: from clinical blood-count diagnostics to neural brain-response visualisers to Ayurvedic AI assistants.
I'm the AI/ML Core at GDG CHARUSAT and Campus Ambassador at Perplexity, where I help students integrate LLM-based tooling into academic workflows. I'm drawn to hard problems in healthcare, education, and automation — places where a good model genuinely changes outcomes.
Deployed on Railway, Render, Netlify, HuggingFace, and Streamlit. Ranked 1644 on LeetCode. NVIDIA Jetson Nano certified. Always writing, always shipping.
Professional
journey.
Full-Stack AI Developer Intern
Hytech EducationDirectly appointed by the Founder & Owner to architect and build an internal GenAI-powered compliance and audit tool. Developed and integrated the complete solution (FastAPI + React) within the scheduled timeframe, which is currently in active, daily internal use.
AI/ML Core
GDG CHARUSATLeading developer student community initiatives in Artificial Intelligence and Machine Learning. Orchestrating workshops, running hands-on hackathons, and mentoring peers in integrating intelligent architectures and neural models.
Campus Ambassador
Perplexity AIDriving adoption of advanced conversational search and large language models across the university ecosystem. Training students and researchers on optimizing prompting methodologies for academic research and productivity.
Projects that
ship.
A2A Firewall
An Inter-Agent Governance Mesh intercepting agent communications. Implements rate limiting, schema verification, permissions matrix, and Groq LLM semantic parsing.
SIGIL
Sovereign Inference Governance & Identity Ledger - replaces embedded AI watermarking with an external, cryptographically signed Merkle-tree attestation ledger. Ed25519 signatures, perceptual hashing, and policy canaries.
ExceptionIQ
AI-powered error intelligence platform built in TypeScript. Detects, classifies, and explains runtime exceptions with contextual precision.
TribeMind
AI app that maps which brain regions activate in response to image, video, or text stimuli — Meta TRIBE v2 model meets Groq-powered LLM neuroscience reports.
Charak Vaidya
Conversational AI assistant grounded in Ayurvedic knowledge — LLM-based Q&A over curated Charaka Samhita for context-aware health guidance.
CBC AI Diagnosis
Clinical-grade ML models for disease prediction from Complete Blood Count reports. Preprocessing, feature engineering and evaluation built for medical accuracy.
Neural ODE - Sepsis
Sepsis progression analysis using Neural Ordinary Differential Equations - continuous-depth models for clinical ICU time-series data.
Metis - AI Hiring Platform
24-hour hackathon build: resumes parsed and fed to a conversational AI interviewer that generates personalised Q&A, scoring, and an HR leaderboard.
Clarix
AI-powered document intelligence - upload any document and get instant, accurate answers through a RAG pipeline with semantic search and LLM-grounded responses.
ExamIDE
Online platform to conduct AI-proctored coding exams - built-in IDE, judge, and session management in one interface.
Traffic Counter CV
Real-time computer vision traffic counter using YOLOv8 ” vehicle detection, counting, and classification from live video streams.
Multimodal RAG System
Production-ready RAG system supporting text, images, code files, and documents - vector embeddings with advanced LLMs for semantic retrieval.
Poker Hand Predictor
ML ensemble model that classifies and predicts poker hand outcomes with high accuracy using advanced feature engineering.
Stock Profit Predictor
Linear Regression-based stock profit forecasting with an interactive JS front-end for real-time portfolio simulation and analysis.
SECOM Yield Optimizer
Semiconductor yield optimization platform built at the IBM Bob AI Hackathon — ML prediction, 120B MoE LLM root-cause analysis, and RL-based recipe tuning with MLflow tracking.
AI Solution Builder
Enterprise AI system that converts business ideas and BRDs into production-ready software — auto-provisions PostgreSQL schemas, REST APIs, UI sandboxes, and CI/CD pipelines via LangGraph agents.
Writing in
public.
Technical deep-dives, architectural decisions, and lessons from building production AI systems.
Why Neural ODEs Beat RNNs for ICU Time-Series Modelling
Continuous-depth models handle irregular sampling and missing data natively — here's how we applied them to sepsis progression and what we learned about clinical deployment.
Building a Merkle-Tree Attestation Ledger for AI Provenance
SIGIL replaces embedded watermarking with external, cryptographically signed provenance — Ed25519 signatures, perceptual hashing, and why this architecture matters for AI governance.
Designing an Agent Firewall: Rate Limits, Schema Verification, and LLM Semantic Parsing
Multi-agent systems need governance layers. A2A Firewall intercepts agent-to-agent communications with a permissions mesh — here's the architecture behind it.
RAG at Scale: From Single-Doc to Multimodal Production Pipelines
Lessons from evolving Clarix into a full multimodal RAG system — chunking strategies, embedding choices, and the retrieval-generation tradeoff in practice.
Let's build something
worth building.
Open to research collaborations, interesting products, and conversations worth having. B.Tech AI & ML @ CHARUSAT · Vadodara, Gujarat.