AI & LLMsBankStatementRAG
Privacy-first offline RAG system for intelligent querying and financial analytics over bank statement PDFs using locally hosted Qwen2.5 7B via Ollama.
Hey there, I’m
Akshat. Agarwal — software developer
I make things for the web.
Sometimes they’re useful.
Sometimes I just want to see what happens.
FIG. 01IDEA → BUILD → REPEAT
Something starts here.Developer. Student. Work in progress.Scroll for the rest of the story
Behind the screen
A person, then a portfolio.
I’m Akshat Agarwal, a software developer and IT student at VIT Vellore. This is where I keep the things I’ve made, what I’m learning, and a few chapters of my journey.
I’m interested in how things work: from the interface you click to the systems behind it. Lately that’s meant building web apps and exploring how AI agents interact with browsers.
There’s always another idea to try.
The experiment shelf
Things I’ve made.
A mix of practical problems, curious ideas, and learning by doing.
Showing 6 projects
AI & LLMs
Full StackThinkBoard
Full-stack collaborative workspace engineered with Next.js, Node.js, Express, MongoDB, and Redis caching with robust token-bucket rate limiting.
Full StackMovie Vault
Modern cinematic exploration web application featuring fluid search, rich metadata filtering, watchlist management, and TMDB integration.
Full StackUnclearTrip
Sleek travel discovery web platform designed to simplify itinerary organization, budget forecasting, and exploration guides.
Systems & AlgorithmsCampus Routing Engine
High-performance campus delivery routing engine built in C implementing Dijkstra's shortest-path algorithm over weighted directed graph representations.
Full StackLineUp
Collaborative task coordination tool offering real-time progress indicators, sprint boards, and priority tagging.
Along the way
A few chapters of work.
The people, teams, and problems I’ve been learning from. Open a chapter for the details.
01Sstudize LabsSoftware Development LeadMar 2026 — Present
Promoted from Software Developer Intern. Leading software architecture and end-to-end technical execution for an AI-enabled EdTech platform serving 500+ students.
- Lead technical direction, feature roadmapping, engineering reviews, and production release cycles across frontend and backend services.
- Architected and shipped full-stack features using React, Next.js, Node.js, Express.js, PostgreSQL, and Supabase across core product surfaces.
- Re-engineered backend workflows and high-traffic APIs for scalability, driving down API response times by 35% on assessment and automated test-generation pipelines.
- Diagnosed and resolved critical production bottlenecks including connection pool exhaustion, request timeouts, and auth rate limits under 1000-VU stress testing.
Next.js · React · Node.js · Express.js · PostgreSQL · Supabase · REST APIs · Stress Testing
02Samsung PRISM — Samsung R&D Institute IndiaR&D Intern (Autonomous Agents)May 2026 — Present
Engineering an autonomous interaction framework utilizing LLM-driven browser agents for complex, dynamic web interfaces and multi-step tasks.
- Developing modular agent orchestration and adaptive interaction pipelines designed to navigate dynamic DOM trees, modals, interstitials, and anti-bot obstacles.
- Integrating browser automation primitives, DOM structure analyzers, LLM multi-step reasoning, and vision-model heuristics to achieve high task success rates.
- Conducting comparative benchmarks between DOM-based and vision-based interaction strategies on shifting web layouts to boost generalizability.
Autonomous Agents · LLMs · Browser Automation · DOM Analysis · Vision Models · Python · Agentic Orchestration
03Metis Intellisystems Pvt. Ltd.AI Developer InternMay 2026 — Jun 2026
Engineered an AI-assisted financial transaction intelligence pipeline processing heterogenous banking records across UPI, IMPS, NEFT, and RTGS.
- Built a deterministic semantic classification engine categorizing transactions across 40+ financial sectors using rule-based signals combined with fine-tuned transformers.
- Boosted classification consistency by 30% through domain-specific feature engineering, lexical normalization, and hybrid classification heuristics.
- Engineered a resilient transaction ingest pipeline that cleans, normalizes, extracts features, and categorizes unstructured transaction narratives.
Transformer Models · NLP · Python · Semantic Classification · Data Normalization · Financial Systems
In the toolbox
Tools for the next idea.
Languages
Java / Python / C / C++ / TypeScript / JavaScript / SQL / HTML & CSS
Frameworks & Frontend
Next.js / React.js / Node.js / Express.js / Tailwind CSS / REST APIs
Databases & Storage
PostgreSQL / MongoDB / Redis / Supabase
AI, Agents & Tooling
Autonomous LLM Agents / Browser Automation / RAG Systems / Local LLMs (Ollama/Qwen) / DOM Analysis & Vision
Infrastructure & DevOps
Docker / Kubernetes / AWS / Git & GitHub / Postman / Vercel
Computer Science Fundamentals
Data Structures & Algorithms / Object-Oriented Programming (OOP) / Database Management Systems (DBMS) / Operating Systems / Computer Networks
Between the finished projects
The work in progress lives on GitHub.
Repos, experiments, and the occasional rabbit hole.
The next chapter
Got an idea? Say hello.
A project, a question, a collaboration, or just a good conversation. My inbox is open.
agarwalakshat2710@gmail.com