FULL STACK / GENERATIVE AIEzResumeA résumé builder that starts with your story.↗Next.js / TypeScript / Gemini / Clerk / Supabase / Monaco / LaTeX
Behind the build +
- Built a Next.js 16 application with Clerk authentication and Supabase persistence for user-specific resume state.
- Designed Gemini-powered generation flows for general, role-based, and job-description-specific resume optimization.
- Added Zod request validation, lightweight rate limiting, retry handling, and strict JSON parsing to harden AI output.
- Generated ATS-friendly LaTeX with escaping safeguards, then exposed a Monaco-based editor and live PDF preview loop.
- Separated prompt rules, content rules, and LaTeX templates so resume behavior can evolve without rewriting the UI.
JD-aware resume generationEditable LaTeX outputAuth + persistence + AI pipeline
Python / Mistral-7B / PEFT / LoRA / bitsandbytes / FAISS / SentenceTransformers
Behind the build +
- Fine-tuned Mistral-7B-Instruct with PEFT/LoRA and 4-bit quantization to make legal-domain adaptation feasible on constrained GPU resources.
- Merged Indian legal QA sources including IPC, CrPC, Constitution QA, and IndicLegalQA-style datasets into a domain-specific instruction corpus.
- Built a FAISS-backed retrieval layer using SentenceTransformers embeddings over Indian-law QA documents.
- Separated model adaptation from retrieval so generated answers can be grounded in indexed legal context instead of relying only on parametric memory.
- Prepared adapter and merged-model paths for flexible inference depending on storage and VRAM constraints.
Domain-adapted Mistral-7BFAISS-grounded legal retrievalLow-resource fine-tuning path
YOLO / PaddleOCR / TensorFlow Lite / Raspberry Pi / Edge TPU
Behind the build +
- Built detection and OCR pipeline with YOLO and PaddleOCR.
- Migrated from Jetson to Raspberry Pi with TPU acceleration.
- Converted PyTorch pipeline to TensorFlow Lite.
- Created augmented dataset with 500% expansion.
- Optimized for edge inference.
~0.9s inference time~60% cost reduction~30% accuracy improvement
