Chemical Engineering @ IIT KanpurAI/MLProcess DesignXR / VRWriter & Poet
Building things that make invisible systems legible — from RAG pipelines that explain every retrieval decision, to VR rooms that teach thermodynamics through memory.
17 projects spanning transparent AI systems, XR learning tools, full-stack applications, and industrial process simulation.
AIRAGLLM Systems
Apr 2026 – May 2026 · Solo · Open Source
01
Traceable Fusion RAG for Algorithms
A transparency-first RAG pipeline that answers algorithm questions grounded in ~1 GB of real open-source code — with every retrieval score, reranking decision, and confidence signal exposed in a live terminal UI.
RAG systems for code are notoriously opaque: users get an answer but have no visibility into why certain documents were retrieved, how confident the system is, or whether the LLM is hallucinating. Existing tools also struggle with multi-language codebases (C++, Python, Java, Markdown) where naïve text splitting destroys function-level semantics.
Most RAG systems give you answers. This one shows you exactly why it chose them.
Jan 2026 – May 2026 + Mar 2026 – May 2026 · Team of 2
02
VR Memory Palace — Chemical Engineering Edition
A Meta Quest XR application encoding 114 Chemical Engineering concepts in a navigable virtual apartment via the Method of Loci — backed by a Spring Boot REST API, PostgreSQL, and a companion spaced-repetition web dashboard.
Undergraduate Chemical Engineering students struggle to retain abstract quantitative concepts using conventional flashcard or lecture review. Existing VR education tools rarely combine an explicit Method-of-Loci spatial encoding protocol with fully interactive VR for abstract technical undergraduate content.
What if your apartment could teach you Chemical Engineering thermodynamics? This one does.
0
Quiz Accuracy Gain
0
Faster Response
0
ChE Concepts
0
API Endpoints
Key Features
›114 ChE concepts spatially anchored to household objects with mnemonic logic
›Meta Quest controller bindings: double-click right trigger reveals floating concept labels and image panels
›Raycasting selection with green/white ray visual feedback
›Spring Boot REST API with 9 endpoints: paginated listing, keyword search, PATCH spaced-repetition scoring
›CSV offline fallback in Unity — activates transparently on network failure
›Web review dashboard with traffic-light Easy/Medium/Hard recall buttons
A production-ready Streamlit dashboard combining Random Forest trading signals with FinBERT news sentiment, Value at Risk, and regulatory compliance monitoring.
One dashboard for trading signals, risk exposure, and regulatory compliance — something most open-source tools keep separate.
›Real-time BUY/SELL signal generation from engineered technical indicators and FinBERT news sentiment
›VaR (Value at Risk) and Sharpe Ratio risk dashboard
›Automated regulatory compliance monitoring with audit checks
A Tkinter desktop AI assistant combining local LLM inference (Ollama) with live multi-source retrieval — Wikipedia, arXiv, news, Wolfram Alpha, YouTube audio, code search — rendered in Markdown with threaded I/O.
A fully local AI assistant that fetches only the sources it needs, when it needs them.
A Streamlit web app that turns text paragraphs into interactive directed concept graphs using TF-IDF, cosine similarity, PageRank, and Dijkstra's shortest path — with physics-based visualisation controls.
Paste any text; see how its ideas connect, which ones matter most, and in what order to read them.
›Paragraph-by-paragraph directed concept graphs from sentence similarity
›PageRank node scoring — most semantically central sentences are largest nodes
›Dijkstra shortest-path optimal reading order table
›Per-graph physics controls: spring length, spring constant, gravity
An 85-dimensional polynomial feature pipeline that models XOR-Arbiter PUF challenge-response behaviour with a linear SVM, achieving 100% accuracy in 0.243 seconds — a 25× speedup over equivalent L1-penalised variants.
Derived a polynomial feature map from hardware security equations; proved a linear classifier is all it takes to break the system — in under a quarter of a second.
›Mathematically derived 85-D feature map from XOR-PUF delay differential equations
A Streamlit web app that fully automates video dubbing into 17 languages — extracting audio, transcribing, translating, synthesising dubbed speech with time-stretching, generating SRT subtitles, and syncing the new audio back.
Upload a video, pick a language — the app handles the rest, including keeping the audio in sync.
›17 supported languages (10 Indian regional languages)
›Silence-based audio segmentation for natural sentence-level dubbing
›Time-stretching preserves original video timing — no sync drift
›Automatic SRT subtitle generation with per-segment timestamps
A minimal, zero-dependency vanilla JS spaced repetition tracker preloaded with 114 Chemical Engineering concepts, designed around a Memory Palace workflow with priority-formula scheduling and JSON export/import.
A zero-install, always-available complement to the VR Memory Palace — review concepts on any device, no headset required.
›114 preloaded Chemical Engineering concepts across thermodynamics, mass balances, phase equilibria
›Memory object mapping (each concept linked to its VR palace anchor object)
A unified theoretical + empirical analysis of oil-price supply shocks across 7 historical episodes, combining AS-AD / IS-LM framework derivations with a Python FRED data pipeline and a fully structured LaTeX research report.
ECO231 Course Project
Mar 2026 – May 2026 · Solo
Python
pandas
NumPy
Matplotlib
FRED API
›9 FRED macroeconomic series: DCOILWTICO, CPIAUCSL, UNRATE, FEDFUNDS, GDPC1
Full industrial process design for Methyl Benzoate via esterification in Aspen Plus: PFR with recycle, 4-column distillation, TEMA shell-and-tube heat exchanger, HEN via pinch analysis, Rankine cogeneration (~16,618 kW), and $64.6M annual profit.
CHE453 Course Project, Prof. Raghavendra Ragipani
Aug 2025 – Dec 2025 · Team of 8
Aspen Plus (UNIQUAC thermodynamics)
Aspen EDR
Fortran
LaTeX
›PFR (L=40m, D=8m, isothermal at 63.4°C): 99.63% overall conversion with recycle
›UNIQUAC thermodynamics validated against NIST Txy data for Water-Methanol binary
PID numerical optimisation achieving IAE=2.4557 on a third-order three-tank cascade — 4× better than best classical PI, with 84% IAE improvement through targeted proportional gain amplification.
MATLAB implementation of McCabe-Thiele graphical distillation design for a multi-section column with side stream withdrawal — VLE curve fitting, operating line construction, reflux optimisation, and automated tray counting.
CHE213 Course Project, Prof. Soumik Das
Mar 2024 – Apr 2024 · Solo
MATLAB
Optimization Toolbox (lsqcurvefit, fsolve)
›VLE fitting: lsqcurvefit on experimental data
›Material balance solved with fsolve for D, W flowrates
Spectrophotometric kinetic study determining that MB photodegradation order shifts from first to zeroth order as H₂O₂ concentration decreases — validated across three experimental scenarios with R² evaluation.
CHE331 Course Project, Prof. Raj Ganesh S. Pala
Oct 2024 – Nov 2024 · Team of 4
MATLAB
Python
UV Spectrophotometry
›Three MB:H₂O₂ molar ratios: 10:1, 29:1, 0 (no H₂O₂)
›Argon lamp photoreactor; UV spectrometer: absorbance measured every minute
Chemical EngineeringProcess Design
17
ChemiEvolve Industries — Feasibility Analysis
Multi-product chemical industry feasibility study designing industrial processes for three chemicals with reactor sizing, economic analysis, and EHS compliance.
Engineer, researcher, poet, and competitive weightlifter — built at the intersection of chemical engineering and AI.
I am a final-year Chemical Engineering undergraduate at IIT Kanpur with a CPI of 9.41/10 — recipient of the Ajai Agarwal Memorial Prize, awarded by the IIT Kanpur Director to the best fourth-year student in Chemical Engineering, and a three-time Academic Excellence Award winner. I grew up in Sambalpur, Odisha, and arrived at IIT Kanpur in 2022 after securing All India Rank 4429 in JEE Advanced among over 200,000 qualified candidates. What followed was four years of building things across an unusually wide range — production-quality RAG systems, Meta Quest VR applications, Spring Boot REST APIs, industrial plant simulations in Aspen Plus, macroeconomic data pipelines, and multilingual video dubbing tools. My flagship project, Traceable Fusion RAG for Algorithms, is a transparency-first retrieval-augmented generation pipeline that indexes roughly 1 GB of open-source algorithm code and exposes every retrieval score, reranking decision, and confidence signal in a live terminal UI. My second flagship, VR Memory Palace — Chemical Engineering Edition, encodes 114 ChE concepts in a navigable Meta Quest apartment using the Method of Loci. A pilot study showed a 7.7 pp gain in quiz accuracy and 60.8% faster response times. Before all of this, I was a Student Mentor Intern at Infinity Learn, where I built 18 functional prototypes and enabled 30+ hackathon teams. Outside engineering, I am a published poet and author — co-published across 10 international literary anthologies, ranked 7th globally at the Storymirror World Writing Championship 2023, and the sole Odia-to-English literary translator in Matruakshar International Journal Vol. I. I also lift weights competitively, having won silver at the Udghosh 2024 under-77 kg category and represented IIT Kanpur at the 56th and 57th Inter-IIT Sports Meets.
Words are the bridges between souls, and through them, we traverse the infinite landscapes of human experience.
