MLH Fellowship Candidate · IIT Jodhpur · CA Intermediate

Building intelligent systems that solve real problems.

I'm Ravi, an Applied AI and Data Science undergraduate at IIT Jodhpur. I build AI-powered applications, agentic systems, retrieval pipelines, and deterministic software tools.

AI Engineering · Machine Learning · Open Source · Software Development

/ 01 About & Engineering Journey

Grounded AI systems, verified by deterministic code.

I am pursuing a B.Sc. in Applied AI and Data Science at IIT Jodhpur (2025–2029) alongside my CA Intermediate (ICAI) qualification. That dual training shapes how I design software: I treat probabilistic language models as orchestrators that must be anchored to deterministic verification layers.

Across my recent builds, that principle shows up everywhere—backing an AI tutor with float64 NumPy invariants in ClaimLab, gating incident remediation behind GRAPH.COPY counterfactual sandboxes in AegisGraph, enforcing AST-allowlisted python -I -S sandboxes with zero false-verified answers in StudyLab, and logging every valuation calculation into an immutable audit trail in FinMod-Agent.

Outside AI pipelines, I also ship full-stack and systems software in TypeScript, React 19, FastAPI, FalkorDB, SQLite/D1, and Rust, with automated CI suites and reproducible local test harnesses.

01 · Agentic Systems & MCP Guardrails

Designing tool-calling agents where safety boundaries live in deterministic tool code and database sandboxes rather than fragile prompt instructions.

02 · Hybrid Retrieval & Grounded Reasoning

Building sparse + dense retrieval pipelines with character-offset citations, epistemic tags, and low-confidence refusal gates.

03 · Quantitative & Financial Engineering

Translating corporate finance, DCF/WACC valuation, and auditing logic into strongly typed, unit-tested Python and Rust applications.

/ 02 Featured Projects

Selected Engineering Systems

All Repositories on GitHub ↗
ClaimLab interactive whiteboard and numerical experiment workspace screenshot 123 Offline + 2 Live Tests · CI Passing

ClaimLab

AI-Powered Teaching & Verified Numerical Learning Workstation

Replaces conversational AI guesswork with verified scientific computing—coupling Meta's muse-spark-1.3 model with a transactional SVG whiteboard, three float64 NumPy labs, and sandboxed paper experiments.

  • LLM Orchestration: Epistemic tagging ([Source], [Experiment], [Background]) and atomic BoardPatch whiteboard state machine.
  • Verified Numerical Workflows: Softmax sensitivity, attention dimension scaling, and causal masking validated within 1e-10 scalar reference tolerance.
  • Adaptive Learning & Security: 2-variant mastery progression, HMAC-SHA256 tenant isolation, CSRF protection, and idempotency fingerprinting.
Python 3.12 FastAPI React 19 TypeScript NumPy float64 pytest
WeMakeDevs × FalkorDB Hackathon

AegisGraph

MCP-Native Graph Incident-Response Agent

Uses FalkorDB to trace what customer PII a CVE actually exposes across code, service mesh, and IAM roles—proves remediation safety in a cloned sandbox graph, then applies surgical, reversible fixes.

  • FalkorDB Multi-Hop Reasoning: Variable-length Cypher traversal across 9 node labels & 11 relationship types with HNSW vector + full-text anchoring.
  • Graph Algorithms & Counterfactuals: Uses algo.betweenness, algo.SPpaths, and GRAPH.COPY + algo.WCC to cut 6 PII attack paths to 0 while keeping 4/4 critical flows intact.
  • MCP Tool Guardrails: 13 stdio MCP tools enforce critical-tier protection and write reversible ActionAudit nodes.
Python 3.12 FalkorDB Cypher MCP FastAPI React
Modular 3-Service Architecture

StudyLab

Modular AI Learning, Hybrid RAG & Verified Solver Platform

NotebookLM-inspired study workstation for AI, data science, and finance learners—combining source ingestion, hybrid cited retrieval, and isolated code-execution solving.

  • Hybrid RAG & Citations: Pairs sparse keyword scoring (exact formulas/tickers) with dense embeddings, reranking, character-offset citations, and low-confidence rejection.
  • AST-Validated Sandbox: Executes Python solver snippets in isolated python -I -S subprocesses with zero filesystem/network escapes.
  • Zero False-Verified Gate: Evaluation suite (run_eval.py) enforces false_verified_rate = 0 across durable SQLite and Docker Compose services.
Python Hybrid RAG SQLite / Postgres AST Sandbox Docker Compose
40 / 40 Pytest Passing

FinMod-Agent (FMVA)

Deterministic Financial Modeling & Valuation Engine

Institutional-grade company valuation engine combining 500+ alias financial normalization, CAPM WACC, Gordon Growth & Exit Multiple DCF, comparable companies, and a 9-screen React UI.

  • Valuation Engines: Full UFCF DCF, live-beta CAPM WACC, peer multiples via yfinance, and Bear/Base/Bull scenario presets.
  • Sensitivity & Auditability: 2D WACC × TGR matrices with monotonicity checks and immutable step-by-step calculation logs.
  • Production Exports & UI: Multi-sheet Excel workbooks, structured JSON API, and 9-screen Vite/React/Recharts workstation.
Python FastAPI Pydantic React Recharts yfinance

More Open-Source Systems on GitHub

ProcureOps AI purchasing and treasury workspace screenshot

ProcureOps AI

Rust (Axum/Tokio) + Next.js + AG Grid procurement cockpit enforcing typed purchase mandates and PayPal Sandbox capture/payout reconciliation.

Rust Next.js PayPal SDK
View Repo ↗

Leadforge Platform

Multi-user career acceleration platform on Cloudflare Workers, D1 SQLite, Drizzle ORM & React 19 with ATS resume builder and Greenhouse/Lever job scraper.

TypeScript Cloudflare D1 React 19
View Repo ↗

AyaWeightDoctor

Zero-gradient rank-1 weight surgery in Layer 16 of CohereLabs/tiny-aya-earth (3.35B), altering 0.6731% of weights in 0.02s to restore alignment without over-refusal.

PyTorch Safety Gradio
View Repo ↗

SQL Query Reviewer

OpenEnv RL environment (GRPO/PPO-ready) with shaped, deterministic per-step rewards across 20 production SQL correctness, indexing, and security tasks.

Python OpenEnv RL Docker
View Repo ↗
/ 03 Architecture & Case Studies

Deep Dive: Problem, Architecture & Verification

Select any flagship project and inspect its engineering tradeoffs, schema/code contracts, and automated test verification.

/ 04 Technical Skills

Categorized Technologies

Every language, framework, and tool listed below is backed directly by shipped code and test suites in my public repositories.

Languages

Python SQL TypeScript JavaScript Rust

Used across ClaimLab, AegisGraph, FinMod-Agent, Leadforge, and ProcureOps AI.

AI / ML

NumPy Pandas SciPy Scikit-learn PyTorch LLMs

Float64 numerical recipes, rank-1 weight surgery, and statistical/macro modeling.

AI Engineering

Hybrid RAG Embeddings Tool Calling MCP Agent Evaluation

13 stdio MCP tools in AegisGraph; sparse+dense retrieval & eval gates in StudyLab.

Backend & Databases

FastAPI REST APIs FalkorDB (Cypher) SQLite PostgreSQL Cloudflare D1

Graph topology + HNSW vector search, SQLite/Postgres migrations, and D1 workspaces.

Frontend

React 19 Next.js Tailwind CSS Vite Recharts

Interactive SVG whiteboard in ClaimLab, 9-screen FMVA cockpit, and AG Grid ledger.

Engineering & Tooling

Git & GitHub GitHub Actions CI Docker pytest Cloudflare Workers

Automated CI workflows, multi-service Docker Compose, and deterministic test suites.

/ 05 Open Source & Achievements

Pull Requests, Engineering Rigor & Credentials

Open-Source Practice & Transparency

Owned OSS Projects vs. Upstream Contributions

I publish my major builds (ClaimLab, AegisGraph, StudyLab, FinMod-Agent) openly under permissive licenses with reproducible setup instructions, architectural docs, and CI checks.

To keep my record strictly accurate: my current public pull request history consists of feature, database migration, and CI-fix PRs across my own and collaborative hackathon repositories rather than third-party upstream repositories.

Why MLH Fellowship: I am applying specifically to collaborate with maintainers and fellows on established open-source ecosystems at production scale.

Verified Certifications, Hackathons & Programs

Anthropic · Jul 2026

AI Fluency: Framework & Foundations

Credential ID: hfgjbcbbuieb
Anthropic · Jul 2026

Introduction to Claude Cowork

Credential ID: e9wrvx2bu8yw
Anthropic · Jul 2026

Claude 101

Credential ID: nrhjymrbxwic
Harvard University · Aspire Institute

Aspire Leaders Programme

Global Leadership & Analytical Cohort
IIT Guwahati · Jan 2026

Convolve 4.0 — A Pan-IIT AI/ML Hackathon

ID: 0587520d-e669-45e4-af1d-90ba7bd35a16
Kaggle (Google) · Dec 2025 & BCG · Oct 2025

5-Day AI Agents Intensive & BCG Strategy Simulation

Multi-Agent Tooling & Strategic Valuation
/ 06 Education

Academic & Professional Foundation

View LinkedIn Profile ↗
2025 — 2029

Indian Institute of Technology (IIT) Jodhpur

B.Sc. in Applied AI and Data Science

Core focus on Python, Machine Learning, Statistical Learning, Data Structures & Algorithms, Linear Algebra, and Optimization. Active in the AI Club and national/global AI engineering hackathons.

Applied AI Data Science Algorithms
2025 — Present (CA Foundation Cleared)

Institute of Chartered Accountants of India (ICAI)

CA Intermediate Candidate

Rigorous domain training in Financial Reporting, Corporate Finance, Costing, Taxation Law, Auditing, and Regulatory Compliance—directly informing deterministic audit & valuation software.

Financial Reporting Auditing Corporate Finance
/ 07 Contact & Resume

Let's build something verifiable.

Open to MLH Fellowship tracks, AI engineering & open-source collaborations, and systems engineering internships.

b25bs1273@iitj.ac.in