# identity.yaml
name : Harshwardhan Tiwari
alias : Eleutherian13 # Greek — "one who brings freedom"
based_in : India 🇮🇳 | Remote-ready
role : AI / ML Engineer · MERN Stack Developer
core_skills :
ml_dl : Python · NumPy · PyTorch · TensorFlow · Keras · Scikit-learn
transformers : Attention mechanisms · Transformer architecture · HuggingFace
rag_llm : LangChain · FAISS · BM25 · Hallucination detection · RAG pipelines
mern_stack : MongoDB · Express.js · React.js · Node.js · REST APIs · JWT
fundamentals : Backprop from scratch · CNNs from papers · ML from linear algebra
currently_learning :
- Computer Vision (YOLO · DETR · OpenCV)
- Vision Transformers (ViT · Swin)
- Multimodal AI systems
- Production MLOps
competition : IIT-R DataForge Top-4 · IIT-R Productathon Top Finisher
streak : 100+ day GitHub streak (and still going 🔥)
honest_note : Some hackathon projects are vibe-coded — that's real!
philosophy : "Never a black box — understand, build, explain."
goal : Contribute to transparent and trustworthy AI
status : Open to collaborations · AI roles · Research"The best way to understand something is to build it from scratch." — Harshwardhan Tiwari
| 💜 | "Any fool can write code that a computer can understand. Good programmers write code that humans can understand." — Martin Fowler |
| 🧠 | "The measure of intelligence is the ability to change." — Albert Einstein |
| 🔥 | "First, solve the problem. Then, write the code." — John Johnson |
| 🚀 | "In theory there is no difference between theory and practice. In practice there is." — Yogi Berra |
| 🌱 | "It's not that I'm so smart, it's just that I stay with problems longer." — Albert Einstein |
I'm not a typical ML engineer who treats models as black boxes. I go deeper — deriving gradients manually, implementing backpropagation from scratch with NumPy, rebuilding legendary CNN architectures from their original papers. Alongside that, I build production-grade full-stack apps with the MERN stack. Some things are vibe-coded and experimental (that's honest), but my core work is always built to be understood and explained.
🔬 Research interest — Hallucination detection · Explainable RAG · Responsible AI
⚙️ Build style — First principles first, then production
🎯 Currently learning — Computer Vision · Vision Transformers · Multimodal AI
🏆 Competition validated — IIT-R DataForge · Productathon Top Finisher
LLM hallucination detection and correction system — my most complete AI project to date. Built a hybrid retrieval pipeline with NLI-based claim verification.
|
A RAG pipeline built around transparency — full source attribution, reasoning chain exposure, and per-answer confidence scoring. No black box outputs.
|
Deep learning with nothing but NumPy. Every forward pass, every backward pass, every optimizer — derived from first principles. Built to understand, not to show off.
|
Classic ML algorithms rebuilt entirely from mathematics. Not because it's faster — because you can't truly understand something you haven't built.
|
Rebuilt landmark CNN architectures directly from their original papers using TensorFlow/Keras — not tutorials, not YouTube walkthroughs.
|
Learning object detection from the ground up — grid cells, anchor boxes, NMS, custom training pipelines. Currently deepening CV knowledge here.
|
|
AI-powered lead management system with scoring, analytics dashboard, and CRM integrations. |
Full production backend — JWT auth, cart, orders, payments — built in a single coding session. |
|
Web-based version with real-time explainability UI. |
My public study journal — annotated notebooks and experiments. |
| 🏅 | Event | Result | Year |
|---|---|---|---|
| 🥇 | IIT-R E-Summit DataForge — Hallucination Detection Track | 4th Place · Judge's Favourite Code · F1 = 0.89+ | 2026 |
| 🏅 | IIT-R Productathon | Top Finisher | 2026 |
| 🚀 | HackNNDD-26 Hackathon | Competed & Shipped | 2026 |
| 🔥 | GitHub Commit Streak | 100+ Days — Unbroken | 2026 |
mindmap
root(("Harshwardhan
2026"))
Research
Hallucination Detection
Explainable RAG
Responsible AI
Currently Learning
Vision Transformers
DETR Object Detection
Multimodal AI
Production MLOps
Building
DL From-Scratch v2
RAG with Full XAI
CNN Paper Implementations
Exploring
Blockchain plus AI
Web3 Integration
AI Ethics
| 🔭 | Implementing now | Vision Transformers (ViT), DETR from papers |
| 🌱 | Actively learning | Multimodal AI, Production ML, Advanced CV |
| 👯 | Open to | Research collabs, AI engineering roles, mentorship |
| 💬 | Ask me about | Explainable AI, RAG, DL from scratch, MERN |
| 🎯 | Long-term goal | Make AI transparent and trustworthy for everyone |
| 📫 | [email protected] | |
| ⚡ | Honest fact | Some of my projects are vibe-coded — and that's okay |
2026 ══════════════════════════════════════════════════════════════
◉ [Q1] 🥇 IIT-R E-Summit DataForge — 4th Place
│ Hallucination Hunter · F1 = 0.89+
│ Hybrid RAG · NLI Verification · FastAPI
│ Singled out by judges for code quality
│
◉ [Q1] 🏅 IIT-R Productathon — Top Finisher
│ Full AI product shipped under competition pressure
│
◉ [Q1] 🔬 Explainable RAG — Deployed
│ Source attribution + reasoning transparency
│
◉ [Q1] 🖼️ CNN Research Papers — Complete
│ LeNet-5 · AlexNet · VGG-16/19
│ Rebuilt from original papers, not tutorials
│
◉ [Q1] 📚 DL + ML From-Scratch Library — Built
│ Neural nets, CNNs, optimizers — pure NumPy
│
◉ [NOW] 🔥 100+ Day GitHub Streak — Unbroken
Consistency over motivation, every day
══════════════════════════════════════════════════════════════════
class HarshwardhanTiwari:
"""
AI / ML Engineer · MERN Developer · First-Principles Learner
"""
stack = ["Python", "ML/DL", "Transformers", "MERN", "Git"]
principles = {
"understanding" : "Deep mastery over surface knowledge",
"transparency" : "Explainable AI > black-box AI",
"first_principles" : "Build from mathematics, zero shortcuts",
"honesty" : "Some things are vibe-coded — and that's fine",
"open_source" : "Share freely — knowledge compounds",
"consistency" : "100+ commit days — habit beats motivation",
}
def approach(self, problem):
steps = [
"Break it to fundamentals",
"Derive the mathematics",
"Implement from scratch",
"Make it explainable",
"Open source it",
]
return f"Solved: {problem!r} — honestly and rigorously"
