AI Fullstack Engineer
Hi, I'm Utkarsh Gupta. I build AI products that hold up in production, from retrieval and evals under the hood to the UI people actually use.
[ OK ] 2,000+ investor queries a day.
[ OK ] Financial data synced for 6,000+ companies.
[ OK ] IPO ingestion live: vendors, web and RHP docs.
✓ All green. 2+ yrs in production.
// Experience
Joined as a Frontend Developer Intern, went full-time, and moved into backend and AI as I took on more ownership.
- Built the RAG and backend behind Ask Iris. It serves 2,000+ queries/day across 500,000+ documents.
- Cut wrong stock links in Ask Iris answers from 4.7% to 0.7%.
- Built evals with LLM-as-a-Judge and deterministic checks, plus a weekly retrieval eval that stays above 90%.
- Merged a dozen document tables into one with dry-run-first backfills and a planned rollback, and revived a vector DB that ran out of disk.
A startup incubated at VIT, run out of the campus incubation centre.
- Built the authentication pages from scratch in React.js for an AI healthcare platform that connects patients with doctors.
// Projects
Ask Iris
An AI research assistant for Indian stocks. It answers questions from annual reports, earnings calls and filings using multi-agent RAG, and cites its sources.
Evals for Ask Iris
An answer-quality suite that mixes LLM-as-a-Judge, rubric and deterministic checks to catch regressions. A weekly retrieval eval scores search against a golden set at top-30 and top-50.
Document Ingestion Pipeline
Company filings come in over a live feed, get converted from PDF to text, then go on to extraction and search. It's built to survive native crashes and dropped connections.
Chatbase: a chatbot for any website
Train a chatbot on any website. It scrapes, chunks and embeds the content, then gives you an embeddable widget with a usage dashboard.
uToob AI: videos into study notes
Paste a YouTube link, get summaries, notes, quizzes and transcript-grounded chat. Bring your own OpenAI or Gemini key.
IPO Data Pipeline
I built the IPO section from scratch. Data comes from vendors, web sources and RHP documents, and a human-in-the-loop upload flow feeds the extracted data into Iris.
Financial Screener
A screener to filter and compare more than stocks: IPOs, ETFs and indices too. Users set their own criteria and scoring to narrow things down.
Company Categorization & Articles
An LLM sorts companies into categories on its own. A second pipeline writes articles straight from company documents.
Sector Analysis Page
A page with sector-level insights, so investors can see trends and comparisons across a whole sector instead of one company.
// Proof
The assistant on this site is a live example of my work. It answers from this portfolio's content, switches to another LLM provider when one fails, and logs every conversation.
// Skills
- Languages TypeScript · Java · Python · JavaScript · SQL
- Backend Node.js · Express.js · REST APIs · FastAPI · Microservices · ETL Pipelines · Cron Jobs
- Generative AI RAG · LLMs · Agentic AI · LangChain · LangGraph · Embeddings · Vector Search · AI Evaluation · LLM-as-a-Judge · Guardrails
- Databases MySQL · PostgreSQL · MongoDB · Redis · Vector Databases
- Frontend React.js · Next.js · HTML · CSS · Tailwind CSS