You will get Clinical RAG Agent — Hybrid Retrieval + Per-Page Citations in 14 Days
Project details
Off-the-shelf RAG demos break on technical content. Layout-heavy PDFs lose structure with naive parsing. Single-vector retrieval misses exact terminology. A single LLM call can't refuse out-of-scope questions.
I'll deliver a production-shaped RAG agent for your corpus — hybrid retrieval, multi-step agent orchestration, per-page citations.
LIVE EXAMPLE
Try clinic-rag.vercel.app on real CDC/USPSTF/NHLBI guidelines (1,614 chunks, 378 pages). 10/10 doc + 9/10 section accuracy on smoke test.
WHAT YOU GET
• Docling layout-aware PDF parsing + semantic chunking
• Hybrid retrieval: pgvector HNSW + tsvector BM25 + RRF + Cohere rerank
• LangGraph agent (triage → retrieve → grade → refine → answer)
• Per-answer citations to source.pdf · page · heading
• Refusal logic for out-of-scope questions
• FastAPI + SSE streaming backend
• Next.js 16 UI with live pipeline trace
• 10-query eval harness scaffolded
• Cloud deploy (Fly + Supabase + Vercel) or your infra
WHY ME
• Shipped clinic-rag.vercel.app solo in 5 days
• Principal Software Engineer — 10 years production Python backend
• Comfortable across pgvector, Pinecone, Weaviate, FAISS, Elasticsearch
I'll deliver a production-shaped RAG agent for your corpus — hybrid retrieval, multi-step agent orchestration, per-page citations.
LIVE EXAMPLE
Try clinic-rag.vercel.app on real CDC/USPSTF/NHLBI guidelines (1,614 chunks, 378 pages). 10/10 doc + 9/10 section accuracy on smoke test.
WHAT YOU GET
• Docling layout-aware PDF parsing + semantic chunking
• Hybrid retrieval: pgvector HNSW + tsvector BM25 + RRF + Cohere rerank
• LangGraph agent (triage → retrieve → grade → refine → answer)
• Per-answer citations to source.pdf · page · heading
• Refusal logic for out-of-scope questions
• FastAPI + SSE streaming backend
• Next.js 16 UI with live pipeline trace
• 10-query eval harness scaffolded
• Cloud deploy (Fly + Supabase + Vercel) or your infra
WHY ME
• Shipped clinic-rag.vercel.app solo in 5 days
• Principal Software Engineer — 10 years production Python backend
• Comfortable across pgvector, Pinecone, Weaviate, FAISS, Elasticsearch
AI Algorithms
Large Language Model, Regression AnalysisAI Applications
AI Chatbot, AI Content Creation, AI-Enhanced Classification, Conversational AI, Natural Language Understanding, Speech SynthesisAI Development Language
PythonAI Tools
Hugging Face, StreamlitAI Models
AlphaCode, ChatGPT, GPT-3What's included
| Service Tiers |
Starter
$1,500
|
Standard
$3,000
|
Advanced
$5,000
|
|---|---|---|---|
| Delivery Time | 5 days | 7 days | 10 days |
Number of Revisions | 2 | 2 | 5 |
AI Model Integration | - | ||
Batch Normalization | - | ||
Database Integration | - | ||
Detailed Code Comments | - | ||
Image Upscaling | - | ||
MLOps | - | ||
Model Deployment | |||
Model Documentation | |||
Model Monitoring | - | ||
Model Testing & Optimization | - | ||
Model Tuning | |||
Natural Language Processing | - | ||
NLP Tokenization | - | ||
Pre-Training | |||
Prompt Engineering | |||
Setup File | |||
Source Code |
Optional add-ons
You can add these on the next page.
Vapi voice mode
(+ 2 Days)
+$800
Larger corpus
(+ 2 Days)
+$1,200
Production observability
(+ 1 Day)
+$70Frequently asked questions
About Waseem
AI Voice Agent Developer | Vapi, Retell, Twilio | LLM and RAG Engineer
Lahore, Pakistan - 12:59 pm local time
I'm focused on AI voice agents (Vapi, Retell, Twilio, Azure Speech) and LLM/RAG applications (OpenAI, Claude, LangChain). I'm currently shipping a set of production voice-agent reference builds — happy to share live demos relevant to your specific use case before you hire.
If you have a job post, send me a sample call or workflow you'd want automated. I'll send back a short Loom showing how I'd build it. No copy-pasted proposal templates.
Stack: Python, FastAPI, Vapi, Retell AI, Twilio, Azure AI Speech, OpenAI, Anthropic, LangChain, Postgres, Redis, AWS, GCP.
Steps for completing your project
After purchasing the project, send requirements so Waseem can start the project.
Delivery time starts when Waseem receives requirements from you.
Waseem works on your project following the steps below.
Revisions may occur after the delivery date.
Corpus Ingestion (Days 1-3)
Set up Docling pipeline. Parse your 5-20 PDFs into layout-aware Markdown. Semantic chunking with metadata preservation. Embed with Voyage AI and index in pgvector HNSW.