Akash Biswas

Generative Ai Engineer with Experience in Rag, Langgraph, Qdrant, Docling @Fraunhofer FOKUS

Kiel, DE
MOBILE NUMBERS
+91 *********19

Signup · Get unlimited contacts

WORK HISTORY

Aug 2024 — Present

Generative Ai Engineer with Experience in Rag, Langgraph, Qdrant, Docling @Fraunhofer FOKUS

View department →

Berlin, DE

Project: Enterprise RAG with Small-to-Big Retrieval & Hybrid Search- Implemented IBM Docling HybridChunker with parent chunks preserving fulldocument context, subdivided into 512-token child chunks with 50-tokenbidirectional overlap, embedded with deepset-mxbai-embed-de-large-v1, deployeddual Qdrant collections for small-to-big retrieval, reducing hallucination from 38% to12% vs baseline fixed-size chunking- Improved recall and precision by 12% & 8%(resp.) compared to dense-only baselineby Hybrid search, combining dense embeddings with BM25 sparse retrieval usingReciprocal Rank Fusion (RRF) weighted 85% semantic / 15% keyword- Built evaluation framework with DeepEval achieving 0.82 answer relevancy, 0.88faithfulness, 0.74 contextual precision using 100 Ragas generated test pairs.Integrated GitLab CI/CD automating regression tests with 7 quality gates, 88% passrate, P95 latency <3.2s, preventing 5 defective deployments.2. Project: German Political Discourse Semantic Evolution Analysis via GottBERTEmbeddings- Extended GottBERT vocabulary with 350+ political terminology tokens, processing193K speeches and 58 manifestos. Applied frequency-weighted aggregation toconsolidate morphological variants (regex-based), capturing 70% of targetdigitalization keywords across both datasets- Developed two-stage neighbor extraction (cosine similarity, regex filtering) yielding15 distinct neighbors with iterative self-variation removal, dual dimensionalityreduction (PCA/UMAP + StandardScaler) revealing temporal semantic evolution- Configured multi-GPU DataParallel inference with three-tier hierarchical caching andsubword reconstruction (BPE, word-level via G-prefix detection), deploying 143Kdata points with automated quality thresholds.

EDUCATION

N/A

Berlin University of Applied Sciences Berlin (BHT)

Master of Science - MS, Data Science

N/A

Deggendorf Institute of Technology

Master of Engineering - MEng, Artificial Intelligence for smart sensors and actuators

N/A

The Scottish Church Collegiate School

Secondary Examination 10th

N/A

Academy of Technology

Bachelor of Technology, Electronics And Communication engineering

N/A

Hare School

Higher Secondary Education - 12th, Statistics - Mathematics - Physics - Chemistry

ABOUT AKASH BISWAS

I\'m a Data Science graduate student and Generative AI specialist passionate about building LLM-powered solutions and advancing open-source AI. From rapid prototyping with Docker and Git to production deployment, I work with teams to deliver impactful results - whether it\'s developing LangGraph agents, implementing graph-based RAG systems, or building transformer applications with LangChain.With 3+ years of hands-on experience and expertise in Python, PyTorch, and Generative AI frameworks, I specialize in LLM fine-tuning with LoRA and QLoRA using distributed parallel computing. I bring both technical depth and practical execution to complex AI projects.I thrive as both an individual contributor and team collaborator, believing that collective effort drives breakthrough innovations in AI.Looking to connect with practitioners and innovators building the next generation of Generative AI solutions.

This profile is compiled from publicly available professional sources. Unifers is not affiliated with or endorsed by LinkedIn. Request removal of this profile.

Akash Biswas — Generative Ai Engineer with Experience in Rag, Langgraph, Qdrant, Docling at Fraunhofer FOKUS in Kiel, DE | Unifers