I build products with AI, and I build AI systems.
I have five years of experience with production software.
Now I make full apps, from the database to the live link, and real people use them.
Currently: I build with AI, and I build AI systems. I live in Delhi, India, and I work remotely.
Open to: Full-time software engineer roles and freelance work. I can build web apps, AI agents, RAG systems and LLM features. I reply in less than 24 hours.
I build full products. I take an idea and make it a live app. I design the database, write the server and the interface, and put it on the internet. Wishcode and GitIssue are live, and people use them. I add security from the start: row-level security, rate limits, encrypted tokens and strict headers.
I build AI systems. I make multi-agent systems with CrewAI, RAG pipelines with ChromaDB, and semantic search with pgvector. I connect LLMs from OpenAI, Anthropic, Groq and Gemini. If the first model fails, a backup model does the work. I also publish tools for AI teams, for example InferenceCI on PyPI.
I ship code that works. I have a background in test automation. Thus, my projects have CI, end-to-end tests and property-based tests from the first day. You can run all the repos below.
| Project | What it is | What to look at |
|---|---|---|
| Wishcode | A public list of app wishes. People post the apps that they want. Developers build them. | Live at wishcode-ai.vercel.app. Next.js 16, React 19, TypeScript and Supabase. pgvector finds wishes with the same meaning, not only the same words. The browser has no database access. All reads and writes go through server functions. The people who want the app vote, and their votes show if the app works. |
| GitIssue | Changes a spoken bug report into a GitHub issue that is ready to post. | Live at report-git-issue.vercel.app. It reads the issue template, the contributing rules and the code of the repo. It uses the real screen names from the code. It finds duplicate issues before you post. Groq writes the draft, and Gemini is the backup. The GitHub token is encrypted with AES-256-GCM. CI runs lint, type checks, unit tests and end-to-end tests on each push. |
| squared-up | Split expenses with friends, and settle with one UPI tap. An India-first Splitwise alternative. | 13.9k lines of code. Django, DRF and Postgres. The money core is plain Python and uses integer paise. It does largest-remainder allocation, debt simplification and UPI intents. Spec vectors and Hypothesis property tests check it. The React and Vite PWA installs on your phone. Run docker compose up to start it. |
InferenceCI (costdiff) |
Finds LLM cost regressions before they ship. On each PR, it runs your scenarios again and compares token spend with main. If the cost increases too much, the check fails. |
Live on PyPI (pip install inferenceci). It reads token usage directly from the OpenAI and Anthropic SDKs through OpenTelemetry. It does not scrape or estimate. An IQR noise floor keeps random variance from hiding real changes. It ships as a GitHub Action. |
| equity-crew | Five CrewAI agents make an equity research report for any listed ticker. | The agents use fundamentals, neural news search and sector peer comparison. I wrote the technical indicators (RSI, MACD, Bollinger, SMA) in pandas, with no TA library. The output is a BUY, HOLD or SELL call with a confidence score, a target price, reasons and risks. A schema checks it, and the report goes to PDF. CI and Docker. |
| git-guide | Multi-agent RAG over the GitLab documentation. CrewAI, ChromaDB and Streamlit. | Async parallel retrieval and smart query routing. Each answer comes from the documentation and shows its source. It runs at zero inference cost. |
| steno | A writing standard that stops AI output from sounding like a language model. | I shipped it and sold it. It is not a demo. It is an instruction layer and a Claude Skills file. It removes the specific signs of AI text: inflated importance, padded clauses, promotional tone, lists of three and a rhythm that never changes. |
| Company | Role |
|---|---|
| Handshake AI | QA Engineer, AI Agent Evaluation (contract) |
| Grey Orange | Software Engineer, QA (SDET) |
| Shway | Software Engineer |
| Yellow.ai | Software Engineer (Customer Success) |
Selected results:
- I made automated regression suites for warehouse and robot workflows. They cut manual test work by 60 to 70%.
- I made an in-house, data-driven PyTest framework. It cut test maintenance by 35%.
- I added end-to-end API contract tests. Backend defects that got to production decreased by 40%.
- I ran JMeter load tests and found the slow paths. Response times at peak load improved by 30%.
- I examined the nightly pipeline runs each day. False positives decreased by 25%.
- I released many iOS and Android features with zero critical defects in production.
B.E. Computer Engineering, SRM Institute of Science and Technology.
Languages · TypeScript · Python · JavaScript · SQL · Bash
Web · Next.js · React · Tailwind CSS · Node.js · Django · DRF · PWA
Data · PostgreSQL · Supabase · pgvector · ChromaDB · MongoDB
AI · OpenAI and Anthropic SDKs · Groq · Gemini · OpenRouter · CrewAI · multi-agent systems · RAG · OpenTelemetry
Platform · Vercel · Docker · Kubernetes · GitHub Actions · Jenkins · Grafana · Linux
Testing · PyTest · Playwright · Selenium · Appium · JMeter · Hypothesis · Postman
All the repos above are open source and free to use. If they help you, please sponsor me. Your support gives me more time to build and maintain them.
Do you have something to build? Send me an email.
raghavg2705@gmail.com