|
Hey, I'm Aasveer 👋, a Computer Science & Engineering (AI/ML) student who enjoys building practical AI applications: systems where intelligent models, backend engineering, data processing and usable interfaces actually fit together. I care less about a model's demo and more about the whole system around it:
|
class Aasveer:
role = "AI/ML Engineer"
location = "Amritsar, India 🇮🇳"
degree = "B.Tech CSE (AI/ML)"
college = "Amritsar Group of Colleges"
focus = [
"Generative AI", "LLMs", "RAG",
"NLP", "Recommendation Systems",
"OCR & Document Intelligence",
]
def philosophy(self):
return ("A model is only one part "
"of a system. The work is "
"making every part fit the "
"person using it.") |
flowchart LR
A["🎯 Understand<br/>the problem"] --> B["📐 Design<br/>the system"]
B --> C["🧪 Build the<br/>pipeline"]
C --> D["🔌 Connect the<br/>components"]
D --> E["🔁 Test &<br/>refine"]
E -. learn .-> A
style A fill:#0b1220,stroke:#22d3ee,color:#e6f6ff
style B fill:#0b1220,stroke:#22d3ee,color:#e6f6ff
style C fill:#0b1220,stroke:#22d3ee,color:#e6f6ff
style D fill:#0b1220,stroke:#22d3ee,color:#e6f6ff
style E fill:#0b1220,stroke:#22d3ee,color:#e6f6ff
| Stage | What it means | Seen in |
|---|---|---|
| 🎯 Understand | Work out what the person actually needs before picking tools | GiftBro AI: the real risk was a model inventing gifts, so ranking real items became the core |
| 📐 Design | Decide the parts, their jobs, and how data moves | PennyBloom: one Python core holds every rule, so the CLI and web app can't disagree |
| 🧪 Build | Build the main path one stage at a time | Wingman AI: OCR → signal analysis → profiling → retrieval, each a separate stage |
| 🔌 Connect | Join backend, data and UI through clear boundaries | InterviewIQ: an Express + Vite server proxies to FastAPI, giving the browser a single origin |
| 🔁 Refine | Try real inputs and plan for missing dependencies | StudyForge AI: without a Gemini key it shows the retrieved source text, clearly labelled |
Six systems. Six different problems. One rule: make AI actually useful, and honest about what it's doing.
| # | Project | What it solves | Core tech | Link |
|---|---|---|---|---|
| 01 | 📚 StudyForge AI | Answers grounded in a student's own PDFs, plus quizzes and planning | RAG FastAPI Gemini |
Repo ↗ |
| 02 | 💰 PennyBloom | Budgets, recurring payments and reports, with one shared core | Python FastAPI React |
Repo ↗ |
| 03 | 🎤 InterviewIQ | Adaptive mock interviews with AI scoring and feedback | React FastAPI Gemini |
Repo ↗ |
| 04 | 💬 Wingman AI | Turns chat screenshots into signals, briefs and reply strategies | OCR NLP RAG |
Repo ↗ |
| 05 | 🎁 GiftBro AI | Ranks real gifts first, then explains them with Gemini | Scikit-learn TF-IDF Gemini |
Repo ↗ |
| 06 | 🎟️ SmartQueue AI | Live digital tokens, staff dispatch and queue analytics | React TypeScript Vite |
Repo ↗ |
💡 Check out the complete interactive showcase, with architectures, evidence and the AI Systems Lab, on my personal portfolio website.
Answers that stay with the source.
Students upload their own PDFs and get answers grounded in that material through RAG, plus Gemini-generated flashcards and quizzes, study modes and planning tools. When there's no API key, a local RAG fallback shows the retrieved source text instead of failing.
PDF ─▶ page-numbered chunks ─▶ TF-IDF + cosine retrieval ─▶ Gemini answer ─▶ flashcards & quizzes
🧑💻 My part: RAG, study planning, quizzes, NLP, FastAPI and SQLite
🛠️ Stack: Python FastAPI Next.js React RAG Gemini API SQLite pytest
🔗 Repository
One core, two interfaces, zero disagreements.
A finance tracker with one Python core and two interfaces: a menu-driven CLI and a React + TypeScript web app over a thin FastAPI layer. It tracks income and expenses, monthly budgets, and recurring payments without duplicates, with monthly reports, category breakdowns and a next-month estimate from a three-month average. Data lives in local JSON with atomic saves, validated CSV import/export and checksummed backups.
transactions ─▶ validation ─▶ budgets & recurring rules ─▶ reports & trends ─▶ atomic JSON storage
🛠️ Stack: Python FastAPI React TypeScript Vite Tailwind CSS Recharts Motion
🔗 Repository
Practice that talks back.
A full-stack mock interview platform for technical, behavioural (STAR) and system-design interviews. Gemini generates adaptive MCQ and open-ended questions, and an AI evaluation step scores every answer on technical depth, communication clarity, problem solving and confidence, followed by strengths, weak spots and a prep roadmap. It also includes a Resume ATS scanner, a job description analyser, a keyword-ranked knowledge base, a voice interview room and a live coding sandbox.
role & skills ─▶ Gemini questions ─▶ answers ─▶ AI evaluation ─▶ feedback & analytics
🛠️ Stack: React 19 TypeScript Tailwind CSS Python FastAPI SQLAlchemy SQLite JWT Gemini API Vite Express
🔗 Repository
From a screenshot to a decision brief.
Reads chat screenshots with OCR and runs a multi-stage pipeline: NLP signal analysis (warmth, reciprocity, engagement), "Conversation DNA" profiling, local RAG retrieval with TF-IDF and cosine similarity, a decision brief and reply strategies. Served through a FastAPI backend with a Streamlit interface and SQLite persistence.
screenshot ─▶ OCR ─▶ NLP signals ─▶ TF-IDF retrieval ─▶ decision brief ─▶ reply strategies
🧑💻 My part: backend, local RAG and NLP, OCR, Reply Lab and Dating Tips
🛠️ Stack: Python FastAPI Streamlit OCR NLP RAG TF-IDF Cosine Similarity SQLite
🔗 Repository
Ranking first, generating second.
A recommender that ranks real catalogue items before any generation happens. Candidates are filtered and scored with TF-IDF, cosine similarity and weighted scoring against the recipient's relationship, occasion, interests, personality, budget and things to avoid. Gemini then explains why each gift fits, but it can never invent a product.
catalogue ─▶ filters ─▶ TF-IDF + signals ─▶ weighted ranking ─▶ Gemini explanation
🛠️ Stack: Python Scikit-learn Pandas TF-IDF FastAPI Streamlit Gemini API SQLite
🔗 Repository
No more standing in line, guessing.
Replaces physical lines with real-time digital tokens. Visitors pick a service, join a standard or priority queue, and follow their live position and estimated wait. Staff open and close counters, call the next token and dispatch people across counters, while rule-based congestion insights and analytics cover waiting and service times, throughput, priority share and no-shows, with CSV/JSON export.
digital token ─▶ queue state ─▶ rule engine ─▶ staff dispatch ─▶ counters & analytics
🛠️ Stack: React TypeScript Vite Tailwind CSS React Context API Lucide React
🔗 Repository
| Capability | Where it's built |
|---|---|
| 🔎 RAG & retrieval | StudyForge AI · Wingman AI · InterviewIQ |
| ✨ Generative AI & LLM apps | InterviewIQ · GiftBro AI · StudyForge AI |
| 🗣️ NLP | Wingman AI · StudyForge AI |
| 🎯 Recommendation engines | GiftBro AI |
| 📄 OCR & document intelligence | Wingman AI |
| 🔌 Backend & API engineering | InterviewIQ · Wingman AI · GiftBro AI · StudyForge AI · PennyBloom |
| 📊 Data & analytics | PennyBloom · SmartQueue AI |
📂 Full skill list (click to expand)
| Area | Skills |
|---|---|
| Languages | Python · SQL · C/C++ · JavaScript · TypeScript |
| AI / ML | Machine Learning · Scikit-learn · Feature Engineering · Recommendation Systems · Forecasting (Prophet) |
| Generative AI & NLP | LLMs · Generative AI · RAG · Prompt Engineering · NLP · TF-IDF · Cosine Similarity · Information Retrieval · AI Output Evaluation |
| AI models & platforms | OpenAI API · Google Gemini API (GenAI SDK) · Anthropic Claude API · Claude Code · Hybrid ML + LLM pipelines |
| Backend & APIs | FastAPI · Flask · Streamlit · REST APIs · JWT Authentication · SQLAlchemy |
| Data & databases | Pandas · NumPy · Matplotlib · Data Analysis · OCR · PDF Parsing (PyPDF2) · SQLite · MySQL · PostgreSQL |
| Frontend | React · Next.js · TypeScript · Vite · Tailwind CSS · HTML · CSS |
| Tools & concepts | Git · GitHub · Jupyter · Google Colab · OOP · DBMS · Data Structures · Problem Solving |
🎓 12 certifications across 6 academies (click to expand)
| Track | Certification | Issuer |
|---|---|---|
| ✨ Generative AI & LLMs | Introduction to Large Language Models | IBM SkillsBuild |
| Introduction to Generative AI | Google Cloud Skills Boost | |
| Introduction to Modern AI | Cisco | |
| AI Practitioner: Ready to Use AI | IBM SkillsBuild | |
| 🤖 Building with Claude | Claude Code 101 | Anthropic Claude Academy |
| Building with the Claude API | Anthropic Claude Academy | |
| Claude Platform 101 | Anthropic Claude Academy | |
| Claude 101 | Anthropic Claude Academy | |
| 📊 ML & data | Machine Learning | IBM SkillsBuild |
| Azure Database for PostgreSQL: Vector Search | Microsoft Learn | |
| 🧭 Prompting & thinking | Introduction to Prompt Engineering with GitHub | Microsoft Learn |
| Critical Thinking in the AI Era | HP LIFE |
+ Going deeper into RAG, LLM evaluation and multi-stage AI pipelines
+ Building AI apps that are reliable, explainable and actually used
+ Sharpening full-stack architecture: clean APIs, clear boundaries
+ Open to AI/ML engineering opportunities and interesting collaborations
🏀 Basketball · 🥊 Boxing · 📖 Novels & folklore · 🎌 Anime · 🎧 Music ·
I'm always up for conversations about AI products, RAG and LLM systems, ML pipelines, full-stack apps
and ideas that might actually work.
