Shubh
Jain
I build intelligent systems that reason — and I race to ship them. Built from first principles.
Engineer.
Researcher.
Builder.
Computer Science at the University of Illinois Urbana-Champaign — a top-5 CS program in the U.S. I design multi-agent architectures, train transformers from first principles, ship full-stack products to production, and lead teams that reach tens of thousands. Equal parts engineer, researcher, and builder.
The season so far.
Solo-built Alpha Engine: deterministic multi-asset signal research across crypto, US & Indian equities, F&O and forex — then backtested it honestly and published the null result rather than a claim.
- Built a deterministic research engine — 22 independent analyzers synthesized into directional signals with calibrated confidence and explicit invalidation prices (33K LOC, solo).
- Integrated 25+ market data sources (Binance, Glassnode, FRED, OANDA, Dhan, AngelOne, NSE, RBI, Finnhub) behind a unified ingestion layer with caching, health tracking and automatic failover.
- Backtested 6,788 signals across 7 assets / 5 years via no-lookahead replay — measured +0.0% edge over base rates and published the null result in FINDINGS.md instead of an unvalidated claim.
- Maintained 1,009 tests across 57 suites; shipped CLI, HTTP API and MCP server interfaces with Docker packaging and daily signal collection via GitHub Actions.
- Designed and built the public organisation site (quanthq.in) solo as a static Astro 5 SSG — type-safe MDX content collections with Zod schemas, interactive canvas visualisations in vanilla script tags, light/dark theming and a contrast audit, deployed to a custom domain via GitHub Pages CI.
LLM refactoring engine for the 25-year-old HDF5 C library with a dual-LLM consensus gate (Claude + GPT-4o both_approve/disagree).
- Built the primary LLM refactoring engine (457 LOC) generating git-format-patch-validated diffs with invariants enforcing no-new-API / no-new-export / no-new-file.
- Designed the multi-model consensus gate — Claude Sonnet and GPT-4o independently review every diff, with split verdicts routed to a human-review disagreements ledger.
- Parallelized the dual-LLM gate (ThreadPoolExecutor, ~2× faster per diff) and added a zero-API-cost faithfulness pre-check catching hallucinated RAG citations in ~1 ms.
- Authored ~280 unit and mocked-E2E tests (14-person repo, 809 total), contributing 80 commits and ~22K lines; prompt caching cut repeat-call cost ~80%.
Owned the data layer of a DistilBERT classifier that detects FDA/FTC advertising compliance violations across 15 regulatory categories.
- Built the preprocessing and tokenization infrastructure turning raw advertising scripts into model-ready tensor datasets (input_ids, attention_mask, labels) in PyTorch and HuggingFace Transformers.
- Expanded and balanced the labelled corpus 5.3× (10 → 53 samples), reaching near-equal class distribution across all 15 violation categories.
- Designed the canonical dataset schema and tensor artifact specification adopted project-wide — the data contract downstream contributors built against.
- Curated a domain-specific corpus spanning social, influencer, e-commerce, broadcast, podcast, email and print advertising so the model generalises across formats rather than one channel.
- Implemented DistilBERT-compatible tokenization across the full 512-token context window with padding, truncation and attention masking.
Led a 31-intern team (Agile) improving a production exam platform.
- Led a cross-functional team of 31 interns with Agile methodology, improving exam model accuracy by 15%.
- Designed data pipelines for large-scale educational datasets and optimization techniques that lifted performance 20%.
- Coordinated multiple bug-detection teams, cutting critical production errors by 40%.
- Awarded Intern of the Month in 3 of 4 months for problem-solving and communication.
Shipped three production AI systems end to end.
- Built and deployed a full-stack AI chatbot with database integration handling 200+ daily inquiries.
- Implemented a BiLSTM misinformation detector at 87% accuracy and a sentiment tool processing 10,000+ reviews.
- Completed a regression price-prediction capstone with <5% mean error.
Selected among 20 students nationwide for an intensive AI/ML program at IIT Delhi's IHFC innovation hub.
- Selected nationwide (top 20) for a hands-on AI/ML program hosted at IIT Delhi's IHFC.
- Built practical models on real datasets — training, evaluation and deployment fundamentals.
- Earned a strong early foundation that directly fed into later research and engineering work.
Things I've shipped at the limit.
Down the pit lane — every build is a race result. Tap any car for the full debrief.
Meter
LLM spend has no enforcement layer — concurrent requests race past the same budget ceiling and streamed responses silently under-bill.
Mnemostack
AI coding assistants retrieve isolated files, losing the dependency chain.
Adaptive Learning Platform
One-size-fits-all learning ignores each student's real mastery frontier.
SIGAIDA Campus Energy
Campus air-quality and environmental data is fragmented and un-forecast.
neuro-rust
Understand backprop by building a neural net with zero ML libraries.
Project Harvest
Dining-hall nutrition data is stale and hard to filter.
Project Helix
Campus events are scattered across 15+ university sources, every one publishing a different and inconsistent HTML structure.
Published from the R&D bay.
UR2PhD Pre-REC Research Training Program
Reproduced a published protein-structure-classification benchmark across 72 datasets in Python/pandas/Matplotlib — finding traditional ML stays competitive with deep learning at a fraction of the runtime — plus a counterfactual name-perturbation replication of an NLP toxicity-bias workflow.
IoT & Precision Agriculture System
Real-time soil / crop / water monitoring with automated irrigation and predictive analytics for sustainable farming in India.
A Comparative Assessment of Advanced Conversational Agents: A Multifaceted Evaluation of ChatGPT, Gemini, Perplexity, and Claude
Compared ChatGPT, Gemini, Perplexity, and Claude using structured factual, open-ended, situational, hypothetical, and creative prompts, with quantitative and qualitative review. The paper reports Gemini higher on factual accuracy and ChatGPT stronger on creative text generation in its examples.
Performance of ChatGPT on Common-Sense Questionnaires
Best Junior Author of the Year ($1,000) · most-downloaded high-school-authored paper in journal history.
Beyond the Black Box: Unlocking Explainable AI
Post-hoc interpretability, counterfactuals & surrogate models applied to healthcare, finance and autonomous systems.
IoT in Agriculture: Revolutionizing Indian Farming
500+ copies distributed to universities; recognized by the Governor of Madhya Pradesh for agricultural innovation.
Calling the strategy.
Races aren't won by drivers alone — 12 crews founded, chaired or run. 200+ people, six cities, tens of thousands reached.
Project Uthaan
FounderBuilt 8 classrooms, installed 100+ computers, educated 2,200+ across digital literacy and vocational programs — 700+ women, 600+ children, 300+ seniors.
AI & STEM Club
FounderEstablished the city's first AI & STEM lab and recording studio; built the school's largest club and won 'Best Club Exhibition' among 50+ clubs.
Model United Nations
LeadershipChaired 8+ committees, ran conferences of 700+ delegates, trained junior secretariats in procedure, motions and conflict resolution.
UIUC Siebel
MentoringOne-on-one support across CS 124/128/173/225 — meeting each student at their exact point of confusion, and advising on RSOs, research access and course sequencing.
SSUAB · UIUC
LeadershipAuthored a comprehensive report on undergraduate research and career-services gaps; mentors peers on résumés, LinkedIn and professional communication.
MLH × Transcend Network
FellowshipRan 24 lean experiments and 15 founder conversations, invalidated 3 core assumptions, and advised founders and Transcend Fund GPs on B2B pricing and AI integration.
UI-CON
EventsRan 20+ panels and 40+ guests across 8 event areas, planning 250+ volunteer hours and troubleshooting 6+ simultaneous rooms live.
Diwali on the Quad
EventsDirected a 4-hour campus festival on a centralized task system — 50+ tracked tasks, 10+ cultural performances, vendor and crowd-flow coordination.
TSVS Student Council
LeadershipRan cybersecurity workshops and a 1,000+ participant TechFest while managing the school website and network — School Colour for ICT, the only high-school recipient.
MetroVaartha
EditorialLed the technology section of a national newspaper: supervised junior editors, edited 200+ articles, and wrote an AI feature that reached 50,000+ readers.
SBI Sustainability Drive
EventsLed an 80-member team with SBI, BHEL and local NGOs across 6 acres of reforestation, including 100+ native species, plus press and community outreach.
Bharat Scouts and Guides
ServiceMentored 400+ underprivileged children, managed 90+ volunteers on community campaigns, and raised $1,500+ for education initiatives.
Designed and ran computer-literacy workshops for underprivileged communities; organized fundraising through handmade product sales.
Ran digital-literacy and computer-basics workshops, improving community access to educational resources.
Coordinated project management, operational strategy and execution for education and community-development initiatives.
What the season won.
A patent, a paid research prize, and a 2.5% founder fellowship — plus the rest of the cabinet.
MLH Commit Fellow
One of 30 fellows in the inaugural founder-track cohort, selected globally alongside Stanford, MIT and Ivy League peers.
IoT & Precision Agriculture Patent
Real-time soil, crop and water monitoring with automated irrigation and predictive analytics for sustainable farming.
Best Junior Author of the Year
For the ChatGPT common-sense study — the most-downloaded high-school-authored paper in the journal's history.
Academic Honours
Sustained academic excellence in a top-5 US Computer Science program, alongside the James Scholar honours program.
High School Overall Topper — Science Batch
Ranked first across the entire science batch in the final school year.
Physics Topper — Grade 12
Highest marks in Physics across the graduating cohort.
School Colour for ICT
The only high-school recipient — awarded for cybersecurity workshops, TechFest and network administration.
Intern of the Month
Three months out of a four-month internship, for problem-solving and communication while leading 31 interns.
Overall Topper — Grade 11
Ranked first among 250 students, achieving 3A — the maximum grade attainable.
Recognition for Agricultural Innovation
For 'IoT in Agriculture' — 500+ copies distributed to universities across India.
Best Club Exhibition
Won against 50+ clubs with an AI-powered chatbot built for school operations.
SilverZone Science Olympiad — International Gold Medalist
100% marks, ranking first internationally among thousands of participants.
Best Delegate
Across 35+ conferences, while chairing 8+ committees and serving twice as Secretary General.
AI/ML Program — Nationwide
Selected among 20 students nationally for an intensive hands-on AI/ML program at IIT Delhi's innovation hub.
Four years, run flat out.
Engineering, research and leadership — running at the same time, not in sequence. Hover or tap a stint to read it.
The build sheet.
Every system on the car, tuned. Drag isn't optional — the whole package has to work together.
Power Unit — Languages
Aero — AI & ML
Chassis — Full-Stack
Telemetry — Data & Infra
Strategy — Research
Comms — Spoken Languages
Let's go racing.
Open to internships, research collaborations, and ambitious builds. Radio in — I usually reply within a day.