Full-Stack & Applied AI Systems Engineer
Software Development Engineer II @ Solfin · Builder of Hortiprise
Gurugram / Delhi NCR, India
I design, ship, and scale production SaaS platforms, autonomous AI agent workflows, and developer tooling that solve real operational problems — not demos.
-
Production FinTech & GenAI @ Solfin (SDE II, Jan 2025 – Present)
- Architected and owned the frontend for Commercial & Industrial (CNI) and Supply Chain Finance (SCF) solar financing journeys in Next.js, React, and TypeScript.
- Engineered an enterprise AI document intelligence microservice (FastAPI + Qdrant + Bedrock/Gemini/OpenAI) with a structured RAG pipeline, raising bill extraction accuracy from 70% → 90%.
- Optimized LLM inference pipelines, reducing latency from 20s → 7s while handling 8,000–10,000 weekly calls at <1% failure rate.
- Built a resource-aware data warehouse job scheduler and an automated Tech-Ops triage platform routing WhatsApp & Gmail tickets to Sales.
-
Vertical SaaS @ Hortiprise (Lead Engineer, 2025 – Present)
- Built an end-to-end nursery management SaaS digitizing inventory, seed germination batches, quotations, billing, and dispatch.
- Engineered a multimodal OCR pipeline (Google Cloud Vision + Tesseract + LLMs) extracting structured data from messy physical invoices.
- Integrated interactive WhatsApp Business API workflows for ordering and dispatch confirmations without requiring an app install.
- Built custom NMS infrastructure monitoring watching GCP VMs, Docker containers, MySQL, and Qdrant to eliminate downtime; automated ~50% of manual operations.
| Project | What It Is | Architecture & Stack |
|---|---|---|
| DiffCommit AI | Published VS Code extension generating semantic git commit messages from staged diffs across 9 AI providers (Claude, OpenAI, Gemini, DeepSeek, Grok, Ollama). | TypeScript, VS Code Extension API, native SecretStorage |
| Syntra | Local-first, privacy-preserving codebase RAG assistant. Indexes code repositories using smart git-diff chunking and SHA hashing to cut re-embedding compute by 90%+. | Python 3.13, Docker, Qdrant, Local Embeddings |
| Unit Atlas | High-precision engineering conversion platform and spatial API with strict schema validation, SlowAPI rate-limiting, and Jest pre-build test gates. | FastAPI, Next.js 16, TypeScript, Jest, Tailwind CSS |
| AI Investment Pipeline | Deterministic 3-stage VC deal sourcing & evaluation engine with on-disk JSON checkpointing for replayability and immutable code rubrics for defensible verdicts. | Python, Multi-LLM Scoring, Pytest, JSON Cache |
- AI & Agentic Systems: Multi-Provider LLM Orchestration (AWS Bedrock, Google Gemini, OpenAI), Structured RAG, Qdrant (Vector DB), Multimodal Vision & OCR (Google Cloud Vision, Tesseract), Autonomous Testing Agents (Appium), Pydantic v2 Schemas, Prompt Injection Defense.
- Backend & Systems: Python (FastAPI, Asyncio, Pytest), TypeScript / Node.js, Relational DBs (PostgreSQL, MySQL / Aiven), Apache Solr (Search & Indexing), Prisma ORM, MongoDB, Redis, Background ETL Schedulers.
- Frontend & UI: Next.js (App Router, Server Components), React, TypeScript, Tailwind CSS, Component Design Systems, State Management & Optimistic UI, Framer Motion.
- Cloud & DevOps: Docker, GCP (Compute Engine, Vision API), Linux/Systemd Administration, GitHub Actions CI/CD, Custom NMS Monitoring.
- Indian Institute of Technology (IIT) Mandi — Minor in Computer Science & Engineering (2024 – 2026) ·
CGPA: 7.0 / 10 - Shri Siddhi Vinayak Group of Institutions, Bareilly (AKTU Lucknow) — B.Tech in CSE (2019 – 2023) ·
CGPA: 8.2 / 10
Built by Vaibhav Agarwal · Powered by GitHub