The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Back to feedPost
Agentic RAG for Document Processing (LangChain, LangGraph, Docling + Qdrant)
I built an end-to-end Agentic RAG pipeline that turns unstructured documents into accurate, context-aware answers. Docling extracts structured content while preserving the original layout, so tables and images stay in place. The content is converted to Markdown, split with recursive chunking into context-preserving chunks, embedded with MPNet, and stored in Qdrant with metadata. LangChain handles retrieval and LangGraph orchestrates the agentic workflow, which fetches the right context for each query using semantic search plus metadata filtering. Redis session memory keeps conversations coherent. Optimizing parsing, chunking, embeddings, and retrieval raised retrieval accuracy from 40–50% to 90–95%.
Document parsing: Docling extracts structured data from documents and keeps tables and images where they belong, instead of displacing them.
Chunking: Content is converted to Markdown with LlamaIndex MarkdownNodeParser, then split with recursive chunking so each chunk keeps its context.
Embedding and storage: The MPNet embedding model creates vectors from the chunks, which are stored in Qdrant along with metadata.
Retrieval: Semantic similarity search is combined with metadata filtering to fetch the most relevant context.
Agent orchestration: LangGraph builds the agentic workflow, with LangChain handling retrieval and LLM calls. The agent decides what to retrieve for each user query.
Memory: Redis-based session memory carries conversational context across the workflow.
Answering: An LLM uses the retrieved context to give grounded, accurate responses.
Tools to tag
Python, LangChain, LangGraph, LlamaIndex, Docling, Qdrant, MPNet Embeddings, Redis, RAG, Agentic AI, LLMs, Document Processing
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Related posts
42% of failed startups built something nobody needed.
That is from one CB Insights analysis of 101 founder post-mortems.
Not bad code. Not bad luck. Nobody needed it.
If you are building an AI product...this is the cheapest fix you will ever make...
Ask these 5 questions BEFORE you build...
1. Who exactly has this problem?
Everyone is not an answer. Name a job title.
2. How are they solving it today?
If they are not solving it at all, it may not hurt enough.
3. What does it cost them?
Hours, money or missed deals. Get a number.
4. Would they pay for a fix this month?
Ask for a pre-order or a pilot, not a compliment.
5. Who have you talked to this week?
Aim for 5 real conversations before writing a line of code.
Copy-paste prompt to prepare your interviews:
I am building [product] for [job title]. Write 10 customer interview questions that uncover how they solve [problem] today, what it costs them, and what they have already paid for. No leading questions.
Compliments do not validate a product.
Payments and pilots do.
Save this before your next build sprint.
Follow me to meet again
Daily learn with me (Maryam Tehsin)
Such an important reminder! 👏 It's easy to focus on building a product, but understanding the customer's real problem should always come first. Real feedback is worth more than compliments!
What if localized videos felt 100% natural with zero distortion? 🚀
Here is a quick test of Zocotev—my in-house Python-based AI video localization tool.
Key highlights of this run:
Zero-shot voice cloning & natural neural voices that match the original speaker's tone and pacing.
Smart Audio Ducking: Automatically suppresses the original background audio to -10% during speech to keep it immersive.
Studio-grade mastering: Normalized to -14 LUFS with precise timeline and pacing synchronization. No hardcoded watermarks or clunky subtitles.
Built completely from scratch in Python. If your agency or production workflow needs a powerful custom video localization pipeline, let's connect!
Dijkstra's 1978 essay on natural language programming describes vibe coding.
He argued that programming needs a formal symbolism because plain language hides mistakes, and that a machine taking instructions in English would make silly errors harder to catch, not easier. Forty-eight years later the machine exists, and the thread about it on Hacker News is split between people who say he's been proven wrong and people who say every failed prompt proves him right.
Both miss where the formal part went. It didn't disappear, it moved. In my own setup the English goes into the brief, and the precision lives in two places I still write by hand: the interfaces, and the tests that have to pass before a change merges. The model fills in the bodies. When the brief is vague the tests catch it, which is the error message Dijkstra wanted instead of a wrong answer.
Where does the precision live in your workflow, the prompt or the tests?