Building and shipping high-performance systems at the intersection of Deep Learning, Production Multimodal AI Systems, Scalable Web Architecture, Enterprise FinTech Platforms and Product Design.
I architect intelligent web applications that bridge the gap between complex data and intuitive user experiences. My work ranges from restoring ancient history through Multi-modal Vision-Language (VL) OCR to building real-time FinTech pipelines. I specialize in end-to-end systems that combine deep learning research with scalable cloud architecture.
I bridge the gap between cutting-edge AI models and real-world client environments. As a Forward Deployed AI Engineer, I specialize in architecting production-grade RAG pipelines, deploying custom data ingestion suites into legacy enterprise workflows, and building high-performance web applications.
- 🔭 Current Focus: Multi-modal Legal AI Inference (Vertex AI / Gemini / Claude), Edge Document Processing, and High-Throughput RAG Systems.
- 🔭 Current Focus: Multi-modal VL OCR for ancient text restoration, Scalable Data Pipelines & Optimizing Chess Engine intuition models.
- 🌱 Learning: Advanced ONNX optimization, AWS Lambda for inference scaling, and fine-tuning Vision-Language Models (VLMs).
- 💼 Business Impact: Monitored & processed 10,000+ case files across 4 law firms, slashing audit times by 90% and saving 90% in operational costs.
- 💬 Ask me about: Forward Deployed AI, Next.js/TypeScript, PyTorch, GCP/Vertex AI, Chess Theory or FinTech/Litigation Funding infrastructure.
| Category | Tools & Technologies |
|---|---|
| AI & LLM Infra | Google Vertex AI (Gemini Pro/Vision), Anthropic Claude API, Multimodal RAG, PyTorch, ONNX |
| Data & Systems | Python (OS Optimization, PyInstaller), Pandas, Data Ingestion Pipelines, Local File Processing |
| UI & Styling | Tailwind CSS, HeroUI, Shadcn UI, Lucide React, Framer Motion |
| Frontend | Next.js, React, TypeScript, PHP |
| Backend & Cloud | GCP (Batch / On-Demand), Vercel Pro (Serverless), Node.js, Flask (Python), NGINX, Laravel |
| Databases | PostgreSQL, Supabase, MongoDB |
| DevOps & Tools | AWS (S3/EC2), Vercel, Docker Compose, GitHub Actions, CLI Tooling |
| Systems & Hardware | C, C++, VHDL, Arduino, Raspberry Pi |
- Forward Deployed AI System: Architected an end-to-end legal inference engine running Gemini Pro and Claude models on GCP, routing requests via hybrid On-Demand and Batch APIs to optimize latency and compute cost.
- Client Integration: Deployed custom Python desktop/CLI document converters across 4 external law firms, interfacing directly with legacy Legal CMS platforms (Proclaim and CaseCube).
- Quantified Impact: Monitored and successfully converted 10,000+ legal cases (20+ file formats to PDF with automated Google Drive sync), reducing case processing time to an average of 10 minutes per file.
- Full-Stack Architecture: Built a central operational mission control and investor-facing platform using Next.js, React, TypeScript, and Supabase (PostgreSQL).
- Long-Running Serverless Compute: Engineered backend API routes on Vercel Pro optimized for long-running execution (up to 800s) to handle complex financial calculations without timeouts.
- Global Compliance & Automation: Programmed automated investor communication engines for 10,000+ users via Mailchimp APIs and built custom UK Holdings tax-statement generator logic for 500+ investors across 40+ countries.
- Data Pipelines: Architected real-time investment dashboards and robust automated CSV/data ingestion workflows for litigation funding using Node.js, PostgreSQL, and Next.js.
- Ancient Text Restoration (UATRIAL): Developed models using Deep Learning and Multi-modal Vision-Language (VL) techniques to restore and attribute ancient epigraphy, merging history with state-of-the-art NLP.
- Neural Chess Engines: Training models on FEN/PGN datasets to predict moves, served via Flask APIs to custom Next.js frontends. Experimenting with Stockfish fine-tuning and custom evaluation functions.
- AI File Converter: Built an end-to-end application that converts various file formats into standardized PDFs, enabling seamless RAG-based inference for AI models.
- AxiaFunder Dashboard: Architected real-time investment dashboards and robust CSV ingestion pipelines for litigation funding using Node.js and PostgreSQL.
- Hack Cambridge Platform: Maintained critical administration and operational platforms for one of the UK’s premier hackathons.
"Build systems that are interactive, intelligent, and scalable. Prioritize clean architecture over quick hacks, and use AI to simplify — not complicate — the user experience."
I’m always open to collaborating on intelligent systems, web platforms, or anything chess-related.
- 🌍 Portfolio: Visit my site
- 💼 LinkedIn: linkedin.com/in/pragash-mohanarajah
- 🔥 Current Streak: 0 days
- 🏆 Longest Streak: 26 days
- ✨ Total Commits: 9,825
- 💖 Commit Breakdown: 562 public (5.7%), 9,263 private (94.3%) · 6,061 owned (61.7%), 3,764 contributed (38.3%)
- 🚀 Repositories: 91 (35 public (38.5%), 56 private (61.5%))
- 👤 Ownership: 87 owned (95.6%), 4 contributed-to (4.4%)
- ⭐ Stars: 39 · 👀 Watchers: 39 · 🍴 Forks: 18 · 🗄️ Archived: 21
- 🧠 Estimated Lines of Code: 1,807,407
- 🤝 Followers: 4 · Following: 11
- 📅 Account age: 2,035 days
C █████████ 37.29% (673,946 LOC)
TypeScript ████ 14.39% (260,089 LOC)
Jupyter Notebook ███ 12.14% (219,406 LOC)
Python ███ 10.61% (191,842 LOC)
HTML ██ 6.33% (114,454 LOC)
JavaScript █ 5.81% (105,089 LOC)
Makefile █ 3.43% (62,016 LOC)
Roff █ 2.99% (54,070 LOC)
JavaScript ████ 17.65% (42 repos)
Python ████ 14.29% (34 repos)
CSS ███ 12.18% (29 repos)
TypeScript ███ 10.08% (24 repos)
HTML ██ 9.24% (22 repos)
Shell ██ 7.14% (17 repos)
Jupyter Notebook █ 5.46% (13 repos)
Dockerfile █ 4.62% (11 repos)
C █████████ 37.29% (33,697,286 bytes)
TypeScript ████ 14.39% (13,004,394 bytes)
Jupyter Notebook ███ 12.14% (10,970,311 bytes)
Python ███ 10.61% (9,592,103 bytes)
HTML ██ 6.33% (5,722,638 bytes)
JavaScript █ 5.81% (5,254,495 bytes)
Other ████████████████ 63.74% (58 repos)
AI / ML █████ 18.68% (17 repos)
Web Apps ███ 10.99% (10 repos)
Data Systems ██ 6.59% (6 repos)
AI / ML ████████████████ 63.46% (1,147,018 LOC)
Other ███████ 27.32% (493,775 LOC)
Web Apps ██ 7.32% (132,237 LOC)
Data Systems 1.90% (34,377 LOC)
JavaScript ████████████ 46.15% (42 repos)
Python █████████ 37.36% (34 repos)
CSS ████████ 31.87% (29 repos)
TypeScript ███████ 26.37% (24 repos)
HTML ██████ 24.18% (22 repos)
Shell █████ 18.68% (17 repos)
Jupyter Notebook ████ 14.29% (13 repos)
Dockerfile ███ 12.09% (11 repos)
Batchfile ██ 6.59% (6 repos)
PLpgSQL █ 5.49% (5 repos)
- Pragash-Mohanarajah/dashboard-axiafunder — AxiaFunder Internal Dashboard Application built with Next.js and Vercel (Fork... (Web Apps · 2005 commits · private)
- AxiaFunder/ai-hdr-inference — Google Vertex AI Inference (AI / ML · 596 commits · private)
- Pragash-Mohanarajah/ai-hdr-inference — AI HDR Case Reviewer Software for Inference from large PDF files (AI / ML · 569 commits · private)
- Pragash-Mohanarajah/exambank-ai-frontend — Exambank AI - Frontend Code Repository (forked from AlphaFactory/exambank-fro... (AI / ML · 474 commits · private)
- Pragash-Mohanarajah/axia-lm-optimizer — Convert to PDF (Other · 393 commits · private)
- AxiaFunder/dashboard-axiafunder (Other · 2060 commits · private)
- AxiaFunder/axiafunder — Monorepo for Axiafunder Applications (Other · 12 commits · private)
Night (00-06) 0.63% (23 commits)
Morning (06-12) ██████████████ 55.04% (1,999 commits)
Afternoon (12-18) ████████ 31.53% (1,145 commits)
Evening (18-24) ███ 12.80% (465 commits)
Sunday ██ 7.51% (762 contributions)
Monday ███ 13.68% (1,387 contributions)
Tuesday ████ 17.97% (1,823 contributions)
Wednesday █████ 19.99% (2,027 contributions)
Thursday ████ 17.33% (1,758 contributions)
Friday ████ 17.11% (1,735 contributions)
Saturday ██ 6.41% (650 contributions)
AxiaFunder/dashboard-axiafunder █████ 20.97% (2,060 commits)
Pragash-Mohanarajah/dashboard-axiafunder █████ 20.41% (2,005 commits)
AxiaFunder/ai-hdr-inference ██ 6.07% (596 commits)
Pragash-Mohanarajah/ai-hdr-inference █ 5.79% (569 commits)
Pragash-Mohanarajah/exambank-ai-frontend █ 4.82% (474 commits)
Pragash-Mohanarajah/axia-lm-optimizer █ 4.00% (393 commits)
Pragash-Mohanarajah/ai-hdr-file-converter █ 3.80% (373 commits)
- AxiaFunder/dashboard-axiafunder — Merge pull request #245 from AxiaFunder/develop
- AxiaFunder/dashboard-axiafunder — Merge pull request #244 from AxiaFunder/taxgenfix3
- AxiaFunder/dashboard-axiafunder — Merge pull request #243 from AxiaFunder/develop
- AxiaFunder/dashboard-axiafunder — Merge pull request #242 from AxiaFunder/fix/tax-statement-generator-data-2
- AxiaFunder/dashboard-axiafunder — Merge pull request #241 from AxiaFunder/taxgenfix3
- Pragash-Mohanarajah/axia-lm-optimizer — feat: improved handoff between file converter and inference codebase by generating csv and jsonl file exports with unknown file metadata
- Pragash-Mohanarajah/axia-lm-optimizer — feat: implement improved file categorisation logic into file converter, only when use_rag flag is enabled; places truly unknown files into unknown folder for downstream AI processing
- Pragash-Mohanarajah/taec-examportal — Fix incorrect backend api call for exports
- Pragash-Mohanarajah/taec-examportal — Fix incorrect backend api call for exports
- Pragash-Mohanarajah/taec-examportal — Allow Superadmins to view all Mark Submission pages and reports
- Pragash-Mohanarajah/taec-examportal — Allow Superadmins to export a summary of all the marks
- Pragash-Mohanarajah/taec-examportal-backend — Add new endpoint to export all mark totals as priveleged user tool for simplicity
- Pragash-Mohanarajah/taec-examportal — Allow Markers to see a summary of the marks, which they have submitted with desensitised information
- Pragash-Mohanarajah/taec-examportal-backend — Add new endpoint to show marks submitted by specific markers on frontend; hide studentId and center number from non-priveleged users
- Pragash-Mohanarajah/Pragash-Mohanarajah — feat: improved outline of current work with mention of legal ai inference
- unoconv/unoserver
- moodle/moodle
- n15hsy/axia-lm
Last updated on Sun, 09 Aug 2026 08:30:08 GMT