
Closed
Posted
I’m building a desktop-first retrieval-augmented generation (RAG) tool and I need a senior Python engineer to take full ownership of the document-ingestion layer. The current prototype already embeds and queries local language models, but the pipeline that feeds them is still thin. Your focus will be to ingest and normalise three document types—plain text, PDFs, and Word files—so they can be chunked, vectorised, and indexed reliably. Here is the landscape you’ll be stepping into: • Stack: Python 3.11, LangChain, FastAPI, SQLite for local mode, Postgres in the cloud, Git for workflow, and a mix of GGUF local models plus an OpenAI fallback. • Environment: Windows is the primary target, so you should be comfortable with PyInstaller or similar packaging to keep the desktop app self-contained. • Goals: clean parsing, robust metadata capture, deduplication on re-ingest, and a schema that stays in sync between SQLite and Postgres. What I will look for first in your message is relevant experience—examples of RAG or vector-based search projects you’ve shipped, any libraries you prefer for PDF or DOCX parsing, and lessons learned from packaging Python on Windows. A concise paragraph linking to repos or write-ups is perfect; no formal proposal deck needed. Acceptance criteria 1. All supported files can be dropped into a watch folder and appear searchable inside the app within 60 s on a typical laptop. 2. Embeddings are stored with full path, hash, and timestamp to prevent duplicates. 3. Switching the persistence backend (SQLite ↔ Postgres) requires only a config flag. 4. The Windows installer bundles every dependency and starts up offline. If this lines up with what you do every day, let’s talk.
Project ID: 40764237
354 proposals
Remote project
Active 10 hours ago
Set your budget and timeframe
Get paid for your work
Outline your proposal
It's free to sign up and bid on jobs
354 freelancers are bidding on average $14 USD/hour for this job

╰┈➤ PYTHON DEVELOPER | RAG & DOCUMENT INGESTION Hi there, I can help strengthen your RAG pipeline with a reliable document-ingestion layer using Python 3.11, LangChain and FastAPI. I’d use PyMuPDF for PDF extraction and python-docx for Word files, with normalised text, metadata capture, content hashing and deduplication to ensure files are processed consistently. I would keep the ingestion layer modular, so switching databases requires only configuration changes. For Windows packaging, I’d also test bundled resources, native dependencies and startup behaviour on a clean machine. ❓ Could you share the current repository or a brief overview of the ingestion code? ❓ Which embedding model and vector index are you using? I would start by reviewing the prototype, then implement and test the ingestion pipeline against your 60-second processing target and offline deployment requirements. Let's discuss this over a quick CALL. Thank you!
$12 USD in 40 days
8.6
8.6

Hello, With over a decade's worth of experience in web, mobile, AI, and automation solutions, I have honed my skills to handle the complexities of your Python RAG Ingestion project seamlessly. My proficiency with Python, a key part of your project stack, is deepened by my knowledge of LangChain and adaptability with FastAPI, an ideal fit for the development journey you're on. Moreover, I've had first-hand experience with SQLite and Postgres, making implementing their interchangeability effortless. Regarding your specific document ingestion requirements (plain text, PDFs, and Word files), I'm well-versed in myriad libraries suited for such tasks, with PyPDF2 and python-docx being top of mind. My adeptness extends to dealing with different platforms too, especially Windows, I am familiar with packaging solutions like PyInstaller that will ensure a user-friendly unpack-and-use option for the end-users. In terms of project goals, I understand the importance of clean parsing, robust metadata capture, deduplication on re-ingest and maintaining a sync between SQLite and Postgres. To meet AC1 expectations, optimizing ingestion efficiency will be a priority for me. AC2 will be met by robustly organizing data using path, hash and timestamps through accurate indexing. AC3 requires meticulous design choices that sync effectively across backends. And lastly, AC4 aligns with my commitment to self-contained deployment which will be key during offlin Thanks!
$30 USD in 27 days
8.0
8.0

Hi, I am Adeel. I built Boostifai, an AI SEO SaaS with Python for the AI work behind a C# and SQL Server core. It pulls in site content, processes it and feeds it to LLMs. I also built Code3Scribe, where the model is held to a strict structured format so it cannot free-write. For your ingestion layer I would hash every file on arrival and key chunks on path, hash and timestamp, so re-ingest skips unchanged files. A watch folder feeds a queue, so parsing never blocks the app. The SQLite and Postgres switch sits behind one repository interface driven by a config flag. I have not packaged a RAG app with PyInstaller before, so I would test the offline Windows bundle early, not last. I can write up the ingestion design and risks for you to keep, whether or not we work together. Which PDFs matter most: text-based ones, or scans that need OCR?
$10 USD in 2 days
8.2
8.2

⭐⭐⭐⭐⭐ Senior Python Engineer for Document‑Ingestion Layer in RAG Tool ❇️ Hi My Friend, I hope you are doing well. I checked all of your project requirements and I can see you are looking for a senior Python engineer to own the document‑ingestion layer. You have no need to look any further as Zohaib is here to help you! My team is already doing 50+ similar projects for RAG and vector search tools. I will build a robust ingestion pipeline that parses txt, PDF and DOCX, captures metadata, deduplicates and syncs SQLite ↔ Postgres. I will also pack the app with PyInstaller so the installer works offline and stays within the same budget. ➡️ Why Me? I can easily do your ingestion layer as I have 5 years of experience in Python, LangChain, FastAPI, SQLite, Postgres, PDF parsing, DOCX handling. Not only this, I have a strong grip on vector databases, metadata schemas and Windows packaging. ➡️ Let's have a quick chat to discuss your project in detail and let me show you samples of my previous work. Looking forward to discussing with you in chat. ➡️ Skills & Experience: ✅ Python 3.11 ✅ LangChain ✅ FastAPI ✅ PDF Parsing ✅ DOCX Parsing ✅ Text Normalisation ✅ Vectorisation ✅ SQLite ✅ Postgres ✅ PyInstaller ✅ Git Workflow ✅ Metadata Management Waiting for your response! Best Regards, Zohaib
$9 USD in 40 days
8.1
8.1

Hi, CnEL India can strengthen your Python RAG ingestion pipeline with reliable document parsing, normalization, deduplication and consistent local/cloud storage. Our methodology: * Audit the existing Python 3.11, LangChain and FastAPI prototype. * Build parsers for TXT, PDF and DOCX using suitable libraries such as pypdf and python-docx. * Normalize extracted text while preserving useful document metadata. * Implement configurable chunking, embedding and indexing workflows. * Track file paths, content hashes and timestamps to prevent duplicate ingestion. * Monitor watch folders and process new or updated files automatically. * Create a shared persistence interface for SQLite and PostgreSQL, switchable through configuration. * Add validation, retries, logging and clear error handling. * Test ingestion speed, retrieval accuracy and repeated-file handling. * Package the Windows application using PyInstaller and validate offline startup. We would first inspect the existing prototype, then implement the ingestion pipeline and verify it against your acceptance criteria before handover. Regards, CnEL India
$12 USD in 80 days
8.1
8.1

Hello, Answers to your questions: 1. Audit: Trace the existing ingestion-to-retrieval flow, inspect parsers, embeddings, schemas, storage, and error handling. Document findings and prioritize fixes before changing code. 2. Versioning: Maintain stable document IDs with separate version records. Keep historical versions traceable while retrieving only active-version chunks by default. 3. Deduplication: Use SHA-256 content hashes to skip unchanged files and detect modifications. Track paths and metadata to handle moves and renames safely. 4. Provenance: Attach document/version IDs, original paths, hashes, chunk indexes, and extraction metadata to every chunk. 5. File watching: Use watchdog for recursive monitoring, backed by reconciliation scans. Handle duplicate events, locked files, restarts, and validate PyInstaller packaging for offline Windows operation. 6. Testing: Test a large mixed corpus with nested folders, malformed files, duplicates, and scanned PDFs. Verify version history, CRUD changes, backend consistency, retrieval quality, recovery, and the 60-second target. My approach is audit first, preserve what works, and make targeted improvements without an unnecessary rewrite. Best, Niral
$15 USD in 40 days
8.1
8.1

Hi, I can take ownership of the Python ingestion layer for your desktop RAG tool. I’ll build reliable parsing for TXT, PDF, and DOCX files, normalize content and metadata, hash documents for safe re-ingestion, and prepare clean chunks for vector indexing. I’ll keep SQLite and Postgres behind the same configuration-based data layer, then test the watch-folder flow against your 60-second target. I’m comfortable working with Python 3.11, LangChain, FastAPI, and Git, and I can handle Windows packaging with PyInstaller so the app runs offline with all required dependencies. I’ll also add clear tests for duplicate files, changed files, backend switching, and installer startup. Which embedding and vector-store interfaces are already implemented, and should the ingestion layer preserve their current schemas? Please start the chat and let’s discuss your current prototype, parser choices, and repository structure.
$15 USD in 20 days
7.7
7.7

Hello I checked your project, and the main focus is building a reliable Python ingestion layer for TXT, PDF, and DOCX files, with metadata, deduplication, embeddings, and SQLite/Postgres support. In the past I built AI document and search systems using Python, LangChain, Pandas, vector indexing, and local model workflows. I’ve also worked with hybrid search using sparse and dense indexes, including Pinecone, where clean document processing and metadata were important. I would build the pipeline around dedicated parsers such as pypdf and python-docx, then normalize content into a shared schema before chunking and embedding. File hashes, source paths, timestamps, and backend configuration would keep re-ingestion safe while allowing SQLite and Postgres to follow the same data model. For the Windows desktop build, I’d test offline startup early and package the Python environment with PyInstaller, including the local model and required dependencies. Could we discuss the current embedding schema and how the watch-folder process is expected to behave when a file is updated or moved? Best regards
$25 USD in 40 days
7.7
7.7

Hello, We carefully studied the description of your project and we can confirm that we understand your needs and are also interested in your project. Our team has the necessary resources to start your project as soon as possible and complete it in a very short time. We are 25 years in this business and our technical specialists have strong experience in Java, JavaScript, Python, Software Architecture, PostgreSQL, SQLite, FastAPI, LangChain and other technologies relevant to your project. Please, review our profile https://www.freelancer.com/u/tangramua where you can find detailed information about our company, our portfolio, and the client's recent reviews. Please contact us via Freelancer Chat to discuss your project in details. Best regards, Sales department Tangram Canada Inc.
$20 USD in 5 days
8.3
8.3

Dedupe on a sha256 of the normalised text, not the filename, so a renamed or re-saved file is caught before it ever reaches the embedder. Days 1-2: PyMuPDF and python-docx parsers plus a watchdog monitor, run on a sample set for you to check. Days 3-7: one SQLAlchemy/Alembic model for both backends, then a PyInstaller build verified on a clean offline Windows VM. 1. Do any PDFs come scanned, so OCR (Tesseract) must be bundled too? 2. Are vectors stored inside the same DB (pgvector / sqlite-vec) or a separate FAISS index? Can start right away and walk through the parser choices in chat.
$12 USD in 7 days
7.9
7.9

Hi,I can help strengthen your RAG document-ingestion pipeline using Python 3.11, LangChain, and FastAPI. I can implement reliable parsing for TXT, PDF, and DOCX files using PyMuPDF and python-docx, with metadata extraction, hash-based deduplication, and consistent indexing across SQLite and PostgreSQL. I’ll also focus on watch-folder ingestion, configurable persistence, and PyInstaller packaging for a self-contained Windows application that supports offline startup. Relevant RAG projects: https://www.freelancer.com/projects/php/Sharepoint-RAG-SQL-GPT-agent/reviews https://www.freelancer.com/projects/php/SQL-RAG-GPT-Agent-with/details I’d be happy to review your current prototype and discuss the best approach to meet your performance and packaging requirements. Thanks.
$12 USD in 40 days
7.8
7.8

Hello, As a seasoned Python engineer at WellSpring Infotech, I’ve built a career on delivering high-quality and efficient solutions using Python, making me the perfect candidate for your RAG Ingestion project. With over 10 years under my belt, I have deep experience in Python 3.11, FastAPI, and Git - all of which feature prominently in your project stack. Additionally, I not only have proficiency working with LangChain and GGUF local models like the prototype you've built but also proficient in PDF and DOCX parsing- essential skills for normalizing the three document types in your tool. In terms of Windows packaging, PyInstaller is something I'm comfortable with, having deployed desktop applications containing similar tech stacks as yours. Furthermore, I'm skilled in PostgreSQL – crucial for the smooth switching of persistence backend as specified in your project goals. I understand that a robust ingestion pipeline is core to the performance of your retrieval augmented generation tool. With a strong track record of creating clean parsers, capturing robust data, implementing metada deduplication measures,censuring schema synchronization between SQLite and Postgres ,and bundling dependencies for offline usability ,I’m more than equipped to exceed all your acceptance criteria and drive excellence in this crucial part of your application. Let's get down to transforming your vision into a functional reality!
$12 USD in 40 days
7.4
7.4

Hello, I can help you with "Python RAG Ingestion Engineer" as per your given set of instructions. Please get in touch to discuss more in detail and start with the initial discusssion. I usually work in small milestones, keep the code organized, and provide regular updates so changes are easy to review. Can you share a few more details? I'll suggest the best implementation and an accurate timeline. ⭐ 5.0/5 from a recent client: "Nice work delievered again. Thank you" Final timeline and cost will be confirmed in chat after a complete understanding and documentation of the project expectations in detail.
$11 USD in 1 day
7.1
7.1

Your weak spot is not retrieval itself but the ingestion layer that determines whether the RAG system stays reliable once real files start flowing in. For this kind of desktop-first setup, I’d focus first on deterministic parsing, content hashing for re-ingest protection, and a storage abstraction so SQLite and Postgres stay aligned behind a config switch. I’ve worked on Python pipelines where PDFs, DOCX, and plain text had to be normalized before chunking and embedding, including metadata capture, duplicate detection, and vector-search readiness. For parsing, I usually evaluate PyMuPDF/pdfplumber for PDFs and python-docx or mammoth depending on how messy the Word source files are; on Windows packaging, PyInstaller works well if dependencies and local model assets are handled deliberately from the start. - Are you already using a watcher library such as watchdog, or should that piece be defined too? - Which embedding store / vector layer is currently in place behind LangChain? - Do you want page-level metadata preserved for PDFs and DOCX, or is document-level metadata enough for v1?
$18 USD in 40 days
7.4
7.4

As an experienced software engineer with a focus on building advanced AI-powered applications, I believe I have the skills you need for your Python RAG Ingestion project. My team at Walking Dreamz Technologies has successfully completed over 300 projects involving complex data extraction and management tasks - skills that are indispensable for your document-ingestion needs. We've also worked significantly with MERN web apps and AI automation, which are relevant to your project's stack and goals. Regarding specific concerns, we're well-versed in Python development, particularly for Windows. Our experience with PyInstaller and similar packaging technologies will ensure that we keep your desktop app self-contained and seamlessly functioning across platforms. Moreover, our wide exposure to different libraries could prove useful as we tackle PDF and DOCX parsing together. Taking a look at some of my past projects, you'll find examples of vector-based search systems I've built that can provide deeper insights into our capabilities in that area. Finally, let's not forget about what's critical beyond the code: our communication style and commitment to long-term quality. At Walking Dreamz Technologies, we prioritize clear and frequent communication through milestones, demos, and progress updates. Furthermore, we understand the importance of clean code and predictable delivery to ensure stable systems that are easily extendible in the future.
$12 USD in 40 days
7.1
7.1

Coming from a strong background in AI and cloud development, I believe I'm well-equipped to take on the challenge of the Python RAG Ingestion Engineer position. Throughout my career, I've consistently designed and developed scalable backend systems, including API integrations, that align perfectly with your project needs. My expertise in Python and FastAPI is complemented by a comprehensive understanding of cloud platforms and database systems. Moreover, my experience building highly intuitive web dashboards for data visualization will be valuable in ensuring smooth indexing and searching of the document types you've mentioned. Additionally, I have hands-on experience working with PDF and DOCX parsing libraries that can streamline ingestion and normalization processes effectively. Lastly, I have considerable experience deploying desktop-first applications using techniques like PyInstaller to cater to Windows users. I assure you that with my technical know-how, your RAG tool will integrate seamlessly with your desired stack and environment. With me onboard, you can expect clean parsing, robust metadata capture, efficient deduplication processes, and a consistent schema between SQLite and Postgres databases. Let's work together to meet your project goals and deliver an exemplary interactive platform.
$20 USD in 40 days
7.2
7.2

Hi, I’m a Python and AI developer with experience working on AI applications and local LLM workflows. I can take ownership of your document-ingestion pipeline, including reliable parsing of TXT, PDF, and DOCX files, text normalization, metadata extraction, deduplication, and chunking for vector indexing. I’d use libraries such as PyMuPDF for PDFs, python-docx for Word files, and LangChain for the ingestion workflow. I’d design a shared persistence layer for SQLite and PostgreSQL, add file hashing and a watch-folder process, and optimize ingestion for your 60-second target. For Windows packaging, I can use PyInstaller and test the bundled application in an offline environment. I’d be happy to review your existing prototype and build on its current architecture while keeping the solution maintainable and reliable.
$12 USD in 40 days
7.2
7.2

Hello, I CAN BUILD A ROBUST, OFFLINE-CAPABLE DOCUMENT INGESTION PIPELINE THAT MAKES YOUR RAG APPLICATION RELIABLE, SEARCHABLE, AND READY FOR WINDOWS DISTRIBUTION. With 12 years of IT development experience, I can take ownership of your Python 3.11 ingestion layer and integrate it with your existing LangChain, FastAPI, and local-model architecture. I will also test re-ingestion, corrupted documents, large PDFs, schema consistency, and retrieval quality. I can review your current repository first and integrate the pipeline without unnecessarily changing your existing prototype. I can share relevant technical examples during our discussion; I will not claim public repositories or shipped projects that I cannot substantiate. Thanks, Christina
$8 USD in 40 days
7.3
7.3

Hi, You need an offline-first Windows RAG ingestion pipeline. The challenge is handling asynchronous file monitoring, clean extraction, preventing duplicate embeddings via hashing, and maintaining hybrid persistence (SQLite/Postgres). Technical approach: - Parsing: PyMuPDF for fast, layout-aware PDF extraction; python-docx for Word. - Pipeline: watchdog triggers FastAPI background tasks. We'll extract SHA-256 hashes, paths, and timestamps before LangChain chunking to skip existing files. - Persistence: SQLAlchemy ORM maintains a unified schema. A config flag swaps SQLite (sqlite-vec) and Postgres (pgvector) dialects. - Packaging: PyInstaller via a custom .spec file. A key lesson in packaging Windows AI apps is managing hidden LangChain dependencies and correctly mapping GGUF binaries in the datas array for offline execution. Relevant systems: - AIDocLink: AI document processing and healthcare vectorization engine. - ORDO AI Platform: Features semantic search, OCR ingestion, and hybrid state sync. Implementation strategy: I recommend building the parsers and ORM layer first, validating the <60s SLA for parsing/chunking. We'll integrate the PyInstaller build step early to test native Windows deployment incrementally, avoiding late-stage dependency bloat. Regards, Rohit
$8 USD in 14 days
7.7
7.7

CHECKED attachments-------------->>Let’s make your ingestion layer the dependable part of the stack, so every dropped file is parsed cleanly, deduplicated, and searchable inside 60 seconds. I have excellent experience building RAG ingestion pipelines with LangChain and FastAPI, using PyMuPDF and python-docx for parsing, hash-based deduplication on re-ingest, and PyInstaller packaging for offline Windows installers. Two questions. Do you want one SQLAlchemy-based schema layer shared between SQLite and Postgres, or are you already committed to a specific migration tool? And which vector store sits behind the index today, such as sqlite-vec, pgvector, or Chroma? Plesae ping to discuss the requirement timeframe and the budget also. Thanks!!!
$12 USD in 40 days
7.1
7.1

Holt summit, United States
Payment method verified
Member since Apr 13, 2025
$15-25 USD / hour
$750-1500 USD
$250-750 USD
£10-15 GBP / hour
$750-1500 USD
₹12500-37500 INR
$250-750 USD
$250-750 USD
$30-250 USD
$8-15 USD / hour
€750-1500 EUR
$30-250 USD
₹1500-12500 INR
$150-200 USD
$15-25 USD / hour
₹250000-500000 INR
$8-15 USD / hour
₹1500-12500 INR
₹100-400 INR / hour
₹1500-12500 INR
$3000-5000 CAD
₹4750-9999 INR