Email • LinkedIn • GitHub • Website

Tech Lead & AI Product Engineer @ Finally (Feb 2024 – Present)
I lead product engineering for AI-driven finance systems—owning PRDs, technical specs, and delivery across a tight, cross-functional team. I design data/ML platforms, productionize LLM/RAG workflows, and ship resilient infra that powers customer-facing products.

• Classify AI (Founding engineer) – Built the flagship AI bookkeeper from the ground up using few-shot classification with Elasticsearch retrieval and a Pinecone RAG store. Implemented LangChain workflows with Redis semantic caching and W&B observability. The system processes 50,000+ transactions/day and cuts manual categorization time by ~80%.
• Merchant enrichment & async scale – Integrated Plaid enriched transactions, merchant grouping optimizations, and Celery task processing; added ML-driven merchant categorization improvements.
• Auto-reconciliation – Built an automated matcher between Plaid-synced bank transactions and accounting entries, reducing month-end close by ~50% and eliminating common bookkeeping errors.
• Bank Aggregator Infrastructure – Architected core infra using Plaid and Teller APIs (OAuth/token mgmt, statement + webhook processing, continuous sync). This backbone powers corporate cards, bookkeeping, and Classify AI v2 with richer metadata.
• Cash-Based Underwriting (Corporate Cards) – Launched a risk system that adjusts credit limits from real-time cash balances & spend patterns, underwriting $3M+ across 50+ companies in 3 months post-release.
• Team leadership – Led a high-velocity team of 3 engineers across the full lifecycle (PRDs → Tech Specs → delivery with OKRs).

• UNAR Labs — ML Engineering Intern
  Built backend systems to enhance multisensory access for visually impaired users using CV + NLP; optimized data pipelines (OpenCV, PyTorch, Transformers, FastAPI) and deployed on GCP with Docker + Hugging Face.

• Outreach — ML Platform Team (Intern)
  Shipped scalable NLP inference on GKE, cutting BERT deployment from 3–4 days to ~30 minutes. Built PySpark + MLflow text pipelines and deployed ONNX models on NVIDIA Triton; delivered Go/Python microservices with CI/CD (CircleCI).

• Mindbowser — Software Engineer (ML)
  Delivered a facial expression recognition system for CRM meeting analysis (VGG-19 transfer learning, ~73% accuracy) with a Flask backend, RBAC/JWT, PostgreSQL + MongoDB GridFS, video ingestion/processing, and REST APIs for jobs/results.

• Gurukul LMS Platform (Django, Next.js, TS, Tailwind, Prisma, Supabase, Docker, Vercel)
  Master’s thesis → IEEE publication. An LLM-enhanced learning environment with RAG + guardrails promoting Socratic problem-solving for DSA. Features a custom code editor, GPT-4 chatbot integration, problem tables, and an intelligent tutoring system.

• Neuralflow — AI-Powered Productivity (Next.js, TS, Clerk Auth, Supabase)
  A focus/workflow app combining Pomodoro with Kanban. Currently adding AI-based ticket grooming and workflow structuring from unstructured input.

• Programming: Python, Go, Snowflake SQL
• Web & Cloud: Django, React, Next.js, TypeScript, TailwindCSS, FastAPI, AWS, PostgreSQL
• ML/AI: PyTorch, Hugging Face, LangChain, LangGraph, Pinecone, FAISS, Weights & Biases
• DevOps & Scale: Docker, Kubernetes, ArgoCD, Celery, Redis, Datadog, Sentry

• Virginia Tech — M.S. (Thesis) in Computer Science, GPA 4.0/4.0
• PICT — B.E. in Computer Science, GPA 8.70/10.00

• Kaggle Expert (Top 1% Notebooks) — Tutorials and competition write-ups
• Active blogger on LLMs, engineering, distributed inference, and more

• 📫 ashwin.rachha@gmail.com
• 💼 linkedin.com/in/ashwinrachha
• 💻 github.com/ashwinrachhavt
• 🌐 ashwinrachhawebsite.vercel.app